181results about How to "Accurate reconstruction" patented technology

The invention relates to the technical field of image processing, and particularly discloses a remote sensing image super-resolution reconstruction method and system based on a prior diffusion model, and the method comprises the steps: carrying out the shallow feature information extraction of a low-resolution image, and obtaining a shallow feature map; the method comprises the following steps: performing multi-scale down-sampling and convolution fusion processing on a shallow feature map and a high-resolution image, and performing high-frequency information extraction and adaptive high-frequency enhancement processing on the processed feature map to obtain condition information; according to the method, deep mining of high-frequency information is carried out on a low-resolution image input as condition information, so that the high-frequency information in the low-resolution image gradually approaches a high-resolution image, and the reconstruction precision is improved; the noise prediction network model guides step-by-step noise removal in the back diffusion process, and the detail reduction capability and precision of natural disaster remote sensing image super-resolution reconstruction are improved.

The invention relates to the technical field of concrete performance simulation, in particular to a concrete crack resistance simulation monitoring method, device and equipment and a storage medium. The method comprises the following steps: acquiring a plurality of groups of internal microstructure images; carrying out different component region segmentation on the plurality of groups of internal microstructure images, and carrying out mechanism structure quantitative analysis to generate mechanism structure parameters of different regions; real-time environmental parameter monitoring is carried out on to-be-monitored concrete based on the multiple sensor nodes, and concrete environmental sensing parameters are obtained; performing environmental stress load excavation on the concrete environmental perception parameters to generate concrete environmental stress load characteristics; and performing multi-frequency filtering noise reduction on the internal three-dimensional microstructure image based on the mechanism structure parameters of different regions, performing regional structure three-dimensional point cloud modeling, and constructing a concrete three-dimensional point cloud model. According to the invention, high-precision and high-efficiency simulation of the crack resistance of the concrete is realized.

The invention discloses a multi-class defect detection method and device based on a variational auto-encoder and a denoising network, and the method comprises the steps: obtaining an industrial product surface defect image, carrying out the preprocessing, and constructing a training set and a verification set; the training process of the method is divided into two stages: in the first stage, a preprocessed training image is input into a variational auto-encoder for training, feature distribution of the image is learned, and features of a normal image are captured by reconstructing the image; in the second stage, a denoising network and a feature guiding module are used for conducting secondary reconstruction on image features in the feature space of the variational auto-encoder, and therefore the reconstruction capacity and the multi-class reconstruction capacity of the model are enhanced; in the reasoning stage, multi-scale features of an original image and a reconstructed image are extracted through a feature extractor, feature comparison is carried out, and an abnormal scoring graph is generated. According to the method, normal information is reserved while the reconstruction capability of the variational auto-encoder is enhanced, and defect missing detection caused by reconstruction failure is effectively reduced.

The invention discloses a remote sensing image super-resolution method, system and device based on implicit neural representation, and belongs to the technical field of deep learning and remote sensing image processing. The method comprises the following steps: firstly, acquiring a data set suitable for a remote sensing image, training a remote sensing image super-resolution model based on the data set, then taking a to-be-reconstructed low-resolution remote sensing image and a pre-constructed query array as input, and performing multi-scale feature extraction by a hierarchical feature pyramid extraction module; the double-path implicit attention feature fusion module carries out fusion, and finally the implicit high-resolution reconstruction module carries out the reconstruction of a high-resolution image. According to the method, the feature pyramid, the attention mechanism and the like are combined with implicit neural representation, the features of different levels are fused together, the problems of multi-scale feature fusion and efficient information extraction in the remote sensing image super-resolution process are solved, the remote sensing image super-resolution precision can be improved, and meanwhile the remote sensing image super-resolution efficiency is improved. And the target of continuous scale super-resolution of the remote sensing image is realized.

The invention provides a large eddy simulation method and system based on a mechanism-only model. The method comprises the following steps: acquiring a current dictionary database of a target turbulent flow field; the current dictionary database comprises multi-dimensional flow field features in a current time step; performing feature screening on the current dictionary database to obtain key flow field features of the target turbulent flow field; inputting the key flow field features into a mechanism-only model, and reconstructing to obtain the sublattice turbulence viscosity of the target turbulence field in the current time step; wherein the mechanism-only model is obtained by training an initial neural network model based on direct numerical simulation data of a sample turbulence field and a sublattice turbulence viscosity label of the sample turbulence field. According to the method, the simulation precision of small-scale turbulence vortex motion in large vortex simulation is improved, so that the problems of poor performance of a bottom layer model and low simulation calculation precision caused by calculation grid scale limitation and sub-grid stress simplification hypothesis are effectively solved.

The invention discloses an intelligent three-dimensional visual perception method based on a parking space scene, and the method comprises the following steps: S1, responding to the illumination change of a parking space environment, and automatically switching a passive or active visual mode; s2, in a passive vision mode, capturing a natural light image by using a binocular camera, and reconstructing a three-dimensional scene; s3, in the active visual mode, projecting a coding pattern, and reconstructing a three-dimensional scene through pattern change; s4, performing semantic segmentation on different objects in the passive visual mode or the three-dimensional parking space scene image in the passive visual mode, and endowing semantic tags; s5, mapping the target object in the two-dimensional image into the three-dimensional scene, and obtaining the position and posture of the target object; by automatically switching the passive vision mode and the active vision mode, the adaptability and the recognition precision of the parking space scene under different illumination conditions are remarkably improved; in a passive vision mode, the system utilizes natural light to capture images, a binocular camera system and a stereoscopic vision technology to accurately reconstruct a three-dimensional scene.

The invention belongs to the technical field of rock core scanning and earth science, and particularly relates to a columnar rock core outer surface hyperspectral image reconstruction method, which comprises the following steps of: 1, acquiring a columnar rock core multi-angle hyperspectral image; 2, cutting an effective area of the multi-angle hyperspectral image of the columnar core; step 3, carrying out radiation correction on the multi-angle hyperspectral image of the columnar rock core; step 4, geometric correction of the multi-angle hyperspectral image of the columnar rock core; 5, splicing the multi-angle hyperspectral images of the columnar rock core; and step 6, removing an overlapping region of the spliced image to obtain a cylindrical core outer surface hyperspectral image reconstruction result. According to the method, radiation brightness distortion and geometric shape distortion generated in the hyperspectral image scanning process can be effectively corrected, and therefore accurate reconstruction of the hyperspectral image on the outer surface of the whole columnar rock core is achieved.

The invention relates to the field of nuclear waste bin transmission image reconstruction. The invention discloses an SA-ResNeXt-based method for reconstructing a nuclear waste bin detection image of an array detector, which can accurately reconstruct a nuclear waste bin chromatography TGS transmission image under sparse transmission scanning. The method comprises the following steps: S1, performing a simulation experiment on the nuclear waste bin based on Geant4 simulation software to obtain training data, and performing a real experiment to obtain verification data; establishing a database for the training data and the verification data; s2, constructing a ResNeXt neural network model according to the transmission image characteristics of the nuclear waste bin; s3, training data in the database are selected to train the ResNeXt neural network model; an optimized ResNeXt neural network model is obtained; and selecting verification data in the database, inputting the optimized ResNeXt neural network model, and realizing reconstruction of the nuclear waste bin transmission image through the optimized ResNeXt neural network model. By adopting the method, the reconstruction accuracy of the TGS image can be greatly improved, and the reconstruction speed can be improved.

The invention belongs to the technical field of three-dimensional three-component velocity measurement, and provides a three-color mask single-color camera particle image velocity measurement particle three-dimensional distribution reconstruction method, which comprises the steps of color particle image shooting, color channel separation, preliminary three-dimensional reconstruction and three-dimensional reconstruction refinement by using a convolutional neural network model. According to the method, three-dimensional reconstruction is performed through the convolutional neural network model, so that the spatial position and the intensity distribution of the particles are more in line with the real situation, the particle elongation effect is greatly reduced, the calculation efficiency is improved, and the method has higher resolution in the depth-of-field direction.

The invention discloses a time-frequency analysis method for processing a multi-component fast-changing time-frequency cross signal, relates to the technical field of signal processing, is suitable for processing and analyzing the fast-changing time-frequency cross signal, and can realize accurate characterization and processing of the fast-changing time-frequency cross signal under a strong noise condition. According to the technical scheme, the method comprises the steps that S1, an EMCT is used for processing collected time-frequency cross signals, and an EMCT result of the signals is obtained; and S2, establishing a new frequency redistribution operator, processing an EMCT result of the signal, obtaining an intermediate frequency track of the signal, and carrying out adaptive time-frequency cross ridge extraction. And S3, establishing a time-frequency coefficient compression method for a time-frequency cross signal, and processing an EMCT result of the signal in combination with the obtained intermediate-frequency trajectory to obtain a time-frequency result. And S4, in the generated time-frequency result, extracting a target component of the signal by using an intermediate-frequency track, and representing and reconstructing a time-frequency cross signal.

The invention relates to the technical field of spine data processing, in particular to a spine feature recognition method and system based on image processing. The method comprises the following steps: acquiring user optical data and spine image data, and performing external form analysis according to the user optical data to obtain external form data; performing spine support analysis on the external form data to obtain spine support data; performing spine skeleton feature extraction according to the spine image data to obtain spine skeleton feature data; and according to the spine support data and the spine skeleton feature data, carrying out spine multi-modal feature fusion to obtain spine feature data so as to carry out spine feature identification auxiliary operation. According to the method, based on multi-modal data fusion, precise spine evaluation and feature recognition are carried out by combining the geometrical morphology of the spine, supporting force analysis and spine image data.

The invention discloses a non-stationary industrial process fault monitoring method based on a self-adjusting non-stationary converter network. The method comprises the following steps: firstly, dividing a variable into a stationary variable and a non-stationary variable by adopting enhanced Dickey-Fuller test; and a sequence stabilization module is designed for non-stationary variables, and the non-stationary variables are converted into stationary variables, so that the reconstruction capability of the model is improved. Meanwhile, a stationary attention removing module is provided to compensate non-stationary information lost by stationary operation, and it is ensured that the model can retain and make full use of inherent non-stationary information when processing non-stationary data. Secondly, a two-stage training strategy based on weight self-adjustment is provided, in the first stage, input window data are reconstructed, the reconstruction deviation of the first stage is used as priori knowledge of attention weight updating of an attention module in the second stage, the model is made to pay more attention to a time sequence with the large deviation, and therefore the convergence speed of the model is increased; and the accuracy and reliability of fault monitoring are improved.

The invention belongs to the technical field of hydrology and landform, and discloses a riverway terrain reconstruction method and device based on energy loss and genetic algorithm coupling, and the method comprises the steps: collecting a remote sensing image of a riverway, the actual measurement data of a hydrological station, and the elevation data; determining a water surface width sequence of a river channel overflowing section based on the river channel water surface vector of the remote sensing image; analyzing actually measured data near the river channel, and determining hydraulic parameters corresponding to the river channel overflowing section; based on the hydraulic parameters, with the minimum energy loss as the optimization target, determining a target width-depth ratio sequence of the river channel overflowing section by adopting a genetic algorithm; based on the water surface width sequence and the target width-depth ratio sequence of the river channel overflowing section, constructing a river channel underwater terrain; and fusing the underwater topography with the elevation model data to complete reconstruction of the river topography. According to the scheme, on the basis of the measured data and the optical remote sensing image of the river channel, various hydraulic parameter characteristics and energy loss corresponding to the overflowing section are considered, and the more accurate river channel terrain can be reconstructed.

The invention discloses a liquid rocket engine turbine pump rotating speed determination method and related device, and the method comprises the steps that a vibration signal and a rotating speed signal of a turbine pump are obtained, and the rotating speed signal is obtained through calculation of an original key phase signal of the turbine pump; processing the vibration signal by combining a Vold-Kalman filter according to the rotating speed signal to obtain a first-order signal of the vibration signal; performing Hilbert demodulation on the first-order signal of the vibration signal to obtain phase information of the first-order signal of the vibration signal; and deriving the phase information of the first-order signal of the vibration signal with respect to time, and calculating the instantaneous rotating speed of the turbine pump according to the derivation result. The invention aims to realize accurate determination of the rotating speed of the turbine pump, so that the accuracy and the reliability of fault diagnosis of the turbine pump are improved.

The invention discloses an image editing method based on a priori constraint inversion algorithm, and the method comprises the steps: firstly inputting an original text, an edited text and a corresponding image by a user, carrying out the preprocessing through a large-scale language model LLM, and obtaining the filtered original text, the edited text and the keyword elements of the edited text in the current edition; secondly, obtaining an edited image through a diffusion model by combining the preprocessed original text and edited text with keyword elements and corresponding images; the method specifically comprises the following steps: inputting a preprocessed original text and a corresponding image into a diffusion model to obtain a submerged space code of a final time step; and inputting the hidden space coding and the preprocessed original text and edit text into a diffusion model to generate a final edit image. In order to enhance the editing effect and utilize the inherent characteristics of the diffusion model, the guiding intensity selector is introduced, and the diversity and the quality of the edited image are improved.

The invention relates to a single-bus current sampling less-phase-shift phase current reconstruction method, which is based on an extended counter electromotive force (EEMF) model, utilizes coordinate transformation to design a current observer on an abc axis, and uses any phase current to estimate other two phase currents, thereby reducing wave phase shift, and reducing electromagnetic noise and power loss generated by the wave phase shift. According to the method, the calculated duty ratio is divided into five states after being centered and aligned, the phase of U, V and W corresponding to an effective current value obtained through sampling is determined according to flag bits VSector and Sshift, and if only one effective current value can be acquired, a corresponding current observer is designed, and three-phase current is reconstructed by adopting a sliding mode observer; and the reconstructed three-phase current can be distributed to U, V and W three-phase driving control circuits of the motor according to the condition of the sector, so that the effect of motor control is realized.

The invention discloses a TEC distribution reconstruction method and device based on a diffusion model, and the method comprises the steps: preprocessing TEC data, constructing a diffusion model, inputting the TEC data into the diffusion model, and carrying out the forward diffusion and reverse denoising training; after training is completed, random noise is input into a diffusion model for reverse denoising to obtain an unknown area image, TEC observation data with missing are input into the diffusion model for forward diffusion, random noise is added to obtain a known area image, and the unknown area image and the known area image are combined to obtain a first intermediate image; and inputting the first intermediate image and the missing TEC observation data into the diffusion model, performing forward diffusion and reverse denoising again, circularly combining the unknown region image and the known region image, and performing iteration for limited times to obtain TEC reconstruction data. According to the method, the TEC data coverage rate is increased, full coverage in the global range is achieved, the TEC data generation precision and reliability are improved, and TEC distribution is accurately reconstructed.

The invention relates to a convenient accompaniment playing method and device based on original sound elimination, a medium and electronic equipment. The method comprises the following steps: carrying out original sound elimination processing on an original audio signal with original sound and accompaniment to obtain an accompaniment signal; according to mixing strategies corresponding to the number of output channels of the playing device, the obtained microphone human voice signals and accompaniment signals are mixed to obtain mixed signals, and the mixing opportunities and mixing modes of the microphone human voice signals and the accompaniment signals in the mixing strategies of different output channel numbers are different; according to the uplink delay duration, the downlink delay duration, the delay output time of the current output of a video picture signal in the original audio and video signal, the audio port output delay, a preset delay interval and the accessory driving delay duration of the last output of the current output, determining the accessory driving delay duration of the current output; and according to the accessory driving delay duration, after the mixed signal is played and output to the playing equipment, carrying out delayed output playing on the video picture signal.

The invention discloses an automatic segmentation and counting method and system for fundus cells, and the method comprises the steps: preprocessing a fundus cell image, inputting the preprocessed fundus cell image into a deep learning model, and enabling a lightweight coupling segmentation head embedded with an attention mechanism module to be introduced into a YOLOv8-seg cell segmentation model, and carrying out the training, thereby obtaining the deep learning model; the attention mechanism module divides the feature map into G sub-features and sends the G sub-features to two parallel branches, one generates a weighted feature map P1 * 1, the other extracts a feature map P3 * 3, 2D global average pooling is used for global space coding of the P3 * 3 and the P1 * 1, and weights w3 * 3 and w1 * 1 are generated; matrix multiplication is carried out on the feature map and the weight, a cross-space interaction weight is generated, and the weight is multiplied by x element by element to obtain Y which belongs to RC * H * W; according to the method, the RGCs in the fundus model can be automatically and accurately circled.

The invention belongs to the technical field of point cloud, and relates to a point cloud attribute coding method based on a graph attention network, which comprises the following steps: acquiring point cloud data, and carrying out block division on the point cloud data; performing detail level generation on the point cloud data subjected to block division, predicting point cloud attributes according to a detail level generation sequence, and inputting a neighbor point set with attribute information into a multi-head image attention layer for feature extraction to obtain a prediction residual error of a current point; encoding the prediction residual error of the current point to obtain an attribute code stream; inputting the attribute code stream into a decoding module to obtain a prediction residual error; searching a neighbor point set of unpredicted points in the prediction point set, and inputting the neighbor point set into the attention layer of the multi-head image to predict the current point; combining the predicted value with the decoded predicted residual error to obtain a reconstruction attribute; according to the method, efficient compression and accurate reconstruction of the point cloud data are realized in a layer-by-layer recursive predictive coding mode, and the method can be applied to a three-dimensional scene needing efficient data transmission and storage.

The invention provides an efficient and stable generative model based on noise scheduling and accelerated sampling, and relates to the technical field of artificial intelligence generative models. According to the efficient and stable generation model based on noise scheduling and accelerated sampling, the distribution proportion between the image and the diffusion factor is controlled, and the growth rate of noise is dynamically adjusted in different stages of the diffusion step, so that the diffusion factor grows faster in the early stage and gradually tends to be stable in the middle and later stages; the influence of too fast noise accumulation on the generation effect is avoided, in addition, the quality of the generated image is optimized by fusing the statistical characteristics of the diffusion factors, and modeling is carried out through the statistical characteristics of the diffusion factors. According to the method, the Gnoise-sigmoid function is used for controlling the increasing rate of the noise in the diffusion process, and the problem of instability of traditional linear or cosine noise distribution in the generation process is solved. Through smoothing noise distribution, rapid diffusion in an early stage and stable transition in a middle and later stage, excessive accumulation of noise is reduced, and the quality of the generated image is improved.

The invention belongs to the technical field of image processing, and particularly relates to a remote sensing image reconstruction model construction and training method and system.The remote sensing image reconstruction model takes prior features as guidance, extracts local features and global features, carries out adaptive feature fusion, reconstructs a low-resolution image, and obtains a high-resolution image. Wherein the prior features provide additional information for the structure, texture, edge and the like of the image, so that the model can reconstruct details of the high-resolution image more accurately; and when the local features and the global features are extracted, the prior features and the global context relationship are fully utilized, and the local features and the global features are fused by adopting adaptive feature fusion processing, so that the performance of the model in a super-resolution reconstruction task is enhanced.

The invention discloses a three-dimensional grid lossless coding and decoding method and device, and the coding method comprises the steps: coding a three-dimensional grid, and obtaining a coding information code stream; decoding the coded information code stream to obtain a reconstructed three-dimensional grid; and comparing the three-dimensional grid with the reconstructed three-dimensional grid to obtain grid difference information, and encoding the grid difference information to obtain a grid difference information code stream. The decoding method comprises the following steps: decoding an encoding information code stream to obtain a reconstructed three-dimensional grid; decoding the grid difference information code stream to obtain grid difference information; and obtaining a decoded three-dimensional grid based on the reconstructed three-dimensional grid and the grid difference information. According to the method disclosed by the invention, complete and lossless grid coding can be realized when any grid containing repeated points, a non-manifold structure, geometric information and attribute information with different connection relationships and the like is coded.

The invention discloses a few-sample fine-grained classification method fusing FRN and Long-LIP models, and the method comprises the steps: collecting fine-grained image data, and making a data set; building an FRN (Field Register Networking) and Long-LIP network model; training the FRN network model, storing the optimal parameters of the FRN network model in the training process, and naming the file of the FRN network model as best.pth; inputting a preprocessed to-be-detected image into the trained FRN network model, inputting the image into the Long-LIP network model to obtain a prediction result, and carrying out weighted summation on the two prediction results. Compared with the prior art, the method has the advantages that deep reconstruction is carried out by adopting the FRN feature map to replace a traditional method for aggregating categories by using similarity average, and an ECA attention mechanism is introduced into the FRN network, so that the extraction capability of the model on key features is improved, and a finer and more accurate prediction result is generated; fRN and Long-LIP models are fused, and a good effect is achieved on a plurality of small-sample fine-grained classification tasks.

The invention discloses a remote sensing image spectral super-resolution reconstruction method, which comprises the following steps: acquiring a remote sensing RGB image data set, and correspondingly converting each remote sensing RGB image into a reference hyperspectral image; the remote sensing RGB image data set is input into the constructed hierarchical semantic perception CNN-Transform model, and a corresponding reconstructed hyperspectral image is generated; constructing a frequency spectrum and space loss function according to the reference hyperspectral image and the reconstructed hyperspectral image, and training the model to obtain a trained hierarchical semantic perception CNN-Transform model; and inputting a remote sensing RGB image to be reconstructed into the trained model to obtain a reconstructed hyperspectral image. According to the method, information loss can be effectively prevented, the stability of the model in different use scenes is improved, and the hyperspectral image is more accurately reconstructed from the remote sensing RGB image.

The invention relates to the technical field of laser gyroscopes, in particular to a dynamic power power supply control system and method for a laser gyroscope, and the system comprises a power supply module, a power collection module and a main control module, the power acquisition module is used for recording and acquiring the output power of a laser of the laser gyroscope, and the main control module performs operation processing based on the actual output power, recorded and acquired by the power acquisition module, of the laser of the laser gyroscope to obtain an adjusting instruction and adjusts power supply parameters of the power supply module; a feedback control unit in a main control module calculates a preliminary adjustment instruction according to a difference value between actual output power and set power, and a feedforward control unit generates an adjustment instruction correction amount according to a laser gyroscope condition parameter after power supply parameter adjustment so as to correct the preliminary adjustment instruction; through the cooperative work strategy of the feedforward control unit and the feedback control unit, the response speed, the control precision and the adaptability of the system are improved.

The invention provides a multi-frame point cloud static scene reconstruction method and device, computer equipment and a medium. The method comprises the steps of obtaining time sequence point cloud data of a vehicle in an automatic driving state; performing frame-by-frame detection on the time sequence point cloud data; performing target tracking on each single-frame scene target to obtain corresponding track attitude data; performing pose estimation on each single-frame scene target to obtain a single-frame estimated pose corresponding to each single-frame scene target; based on the track attitude data corresponding to each single-frame scene target and the single-frame estimated pose corresponding to each single-frame scene target, carrying out point cloud separation on the single-frame scene target belonging to the dynamic target to obtain a plurality of single-frame static point cloud data; and splicing a plurality of single-frame static point cloud data based on the single-frame estimated pose corresponding to each single-frame scene target so as to reconstruct a static point cloud scene which does not contain the dynamic obstacle. Therefore, accurate static three-dimensional scene reconstruction is effectively realized.

The invention discloses a code multiplexing method applied to reading of an array type plastic scintillator detector, and relates to the technical field of particle detection, and the method comprises the steps: designing the section shape, arrangement mode and number of each plastic flash strip in a one-dimensional plastic flash strip array; the requirement that at least two plastic scintillation strips with continuous hitting positions and continuous numbers are bound to be hit when particles to be detected hit the array type plastic scintillator detector is met; according to the section shape, the arrangement mode and the number of each plastic flash strip, determining coding parameters based on a detector channel and a signal reading channel; generating a coding table according to the coding parameters; and establishing a detection reading framework according to the coding table, performing particle detection on the to-be-detected particles, and reversely deducing the hit detector channel according to the serial number corresponding to the signal reading channel of the signal received in the detection process. According to the invention, the number of required signal reading channels can be reduced, and the complexity and cost of an electronics system are reduced.

The invention relates to the technical field of current signal processing, in particular to a current signal compression collection reconstruction method and device, a storage medium and electronic equipment, and the method comprises the steps: carrying out the sliding collection of a current signal through a flexible Blackman window, and obtaining a sliding collection signal; constructing a partial orthogonal Gaussian measurement matrix; performing compression acquisition on the sliding acquisition signal by using the partial orthogonal Gaussian measurement matrix to obtain a compressed signal; carrying out sparse transformation on the sliding acquisition signal, and solving to obtain a frequency domain transformation matrix; calculating by using the frequency domain transformation matrix and the partial orthogonal Gaussian measurement matrix to obtain a sensing matrix, and iteratively solving a sparse signal by using an algorithm according to the compressed signal and the sensing matrix; and according to the frequency domain transformation matrix and the sparse signal, calculation is carried out to obtain reconstruction data. By means of compression acquisition and reconstruction, low-volume data acquisition is achieved, high-frequency fault arc characteristics in data are effectively reserved, and the problem of fault arc characteristic deficiency after traditional current signal acquisition is solved.

The invention provides a speech synthesis model training method, a speech synthesis method, a speech synthesis device and speech synthesis equipment, and belongs to the technical field of computers. The first speech synthesis model comprises a language network and a diffusion network; the method comprises: processing a sample text and a sample audio through a language network to obtain a first audio feature, the sample text and the sample audio being consistent in content, the first audio feature comprising a content feature of the sample text and an acoustic feature of the sample audio; performing noise prediction on the sample audio added with the reference noise based on the first audio feature through a diffusion network to obtain predicted noise; determining a first training loss based on the reference noise and the predicted noise; and updating parameters of a language network and a diffusion network in the first speech synthesis model based on the first training loss to obtain a second speech synthesis model. According to the technical scheme, the purpose of end-to-end training of the speech synthesis model is achieved, the training steps are simplified, and the training efficiency of the model is improved.