24results about How to "Guaranteed naturalness" patented technology

The invention relates to the technical field of signal processing, in particular to an unmanned aerial vehicle self-noise filtering method and system based on multichannel time-frequency spatial filtering. The method comprises the following steps: acquiring a multi-channel noise-containing sound signal by using a microphone array carried by an unmanned aerial vehicle, and performing short-time Fourier transform to obtain a frequency domain signal; estimating a ratio mask by using a deep neural network, and performing logarithm normalization processing on the frequency domain signal; remodeling the sound signal to match the FFT length of the time-frequency spatial filter, and constructing the time-frequency spatial filter in combination with the multi-channel ratio mask; removing the self-noise of the unmanned aerial vehicle by using the filter and obtaining a target sound signal; and finally, converting the frequency domain signal into a time domain signal through inverse short-time Fourier transform. Through innovative fusion of multi-channel acquisition, deep learning and time-frequency space filtering, the self-noise of the unmanned aerial vehicle in a complex environment is accurately filtered, and the accuracy and efficiency of transformer substation noise monitoring are remarkably improved. Reliable acoustic data are provided for state monitoring of the power equipment, and the method has important engineering application value.

According to the character expression driving method and device, the storage medium and the computer equipment provided by the invention, during expression driving, the continuous video frame images and the expression driving target in the character video are determined firstly, so that expression driving is realized on the video frame images, and the naturalness of an expression driving effect is ensured; and then face alignment is carried out based on the face key points in the video frame image to obtain a standard face image, expression driving is carried out on the standard face image based on the expression driving target through the target expression driving model, an expression driving video is generated, and thus the quality of the expression driving effect is improved. The target expression driving model is formed by the three-dimensional convolutional neural network and is used for mapping the two-dimensional image to the three-dimensional hidden space for feature coding and expression driving, so that the model can accurately track human faces at various angles in the video frame image and perform expression editing and driving; and expression natural connection of video frame images between frames is realized under the condition that the quality of a figure video is not influenced.

The invention provides a spinal endoscopic image deblurring method and system, and the method comprises the steps: carrying out the multi-scale convolution operation of a spinal endoscopic image, obtaining the difference of image contrast among different scales, segmenting the spinal endoscopic image into a plurality of local structure images, and obtaining a plurality of local structure images; determining the brightness uniformity of each local structure image through the distribution gradient of the pixels in each local structure image and the difference quantity of the contrast; determining the dependency relationship of the textures between the adjacent local structure images through the texture difference between the adjacent local structure images on the spine tissue structure; performing contrast enhancement on the brightness of the local structure image according to the dependency relationship and the brightness uniformity to obtain a local enhanced image; and fusing and reconstructing the fuzzy region through all the local enhanced images to obtain a deblurred spinal endoscope image. By adopting the scheme provided by the invention, local adaptive equalization enhancement of the image brightness of the spinal endoscope can be realized, so that detail blurring caused by non-uniform illumination is reduced.

The invention relates to the technical field of color correction, in particular to a baked porcelain false tooth color adjusting method and system based on image analysis, and the method comprises the following steps: identifying teeth, gingiva and surrounding tissues in an image, determining a region boundary by using an edge detection technology, and calibrating a region according to gray difference. According to the method, teeth and surrounding tissues are accurately segmented by using an edge detection technology, accurate identification of a tooth area in an image is realized, a clear starting point is provided for color adjustment, a wave band with high reflectivity is selected, the quality of image data is optimized, the precision of subsequent color analysis is ensured, and hue, brightness and saturation are adjusted in a CIELAB color space; the matching degree of denture colors and natural teeth is improved, the visual coherence is enhanced, neural network analysis and time sequence data are integrated to predict long-term color changes, a real-time feedback mechanism automatically adjusts color parameters according to illumination conditions, the visual effect is continuously optimized, and the naturalness and consistency of colors in different environments are guaranteed.

The invention discloses a speaking face video generation method based on Gaussian splashing, and mainly relates to two stages. In the first stage, a large-scale data set is used for pre-training a model for generating dynamic facial expression coefficients based on audio. The model learns universal facial expressions and motion modes based on audio information and image information, and can generate dynamic facial expression coefficients synchronized with audio. And a second stage, performing fine adjustment on the pre-training model by using the data set of the target person, obtaining a face expression coefficient based on the fine-adjusted model after fine adjustment is completed, further training a Gaussian deformation network driven based on the face expression coefficient to perform deformation processing on the Gaussian point cloud, and performing rasterization to obtain a speaking image of the target person. And finally, integrating the two-stage model, and synchronously merging the frame-by-frame image and audio content to realize a speaking face video generation task. According to the method, the dynamic face corresponding to the brand new audio which does not appear in the data set can be quickly generated, the naturalness and high quality of the generation result can be ensured while the high generation speed is kept, and audio and video synchronization and high reduction of expression details are realized.

The invention belongs to the technical field of image data processing, and particularly relates to an editable face three-dimensional reconstruction method and system and a storage medium. The method mainly comprises the following steps: acquiring expression-free image data and expression image data of a modeler; obtaining a neutral model and a real expression model; training a deformation field, a color field and a super-resolution neural network; training an expression encoder to construct an expression base vector; and forming a facial three-dimensional model with expressions of the modeler through the expression base vector and the finally generated model, and outputting a rendered image. The invention further provides a system for implementing the method. Face three-dimensional reconstruction is carried out based on the neural radiation field technology, an accurate, natural, attractive and real face three-dimensional model can be generated, flexible and convenient personalized adjustment can be carried out, and the method has good application prospects in the fields of medical treatment and medical beauty.

The invention provides a real-time call translation method and device, relates to the technical field of data processing, and is applied to a Bluetooth headset, and the Bluetooth headset establishes communication connection with a mobile terminal through a Bluetooth hands-free protocol and establishes communication connection with a translation application running in the mobile terminal through a Bluetooth low-power-consumption general attribute protocol. The method comprises the following steps: acquiring an original sound to be translated, and performing audio compression coding on the original sound to be translated to obtain a compressed audio; transmitting the compressed audio to the translation application through a Bluetooth low-power-consumption general attribute protocol; receiving a translated audio returned by the translation application through a Bluetooth low-power universal attribute protocol, wherein the translated audio is an audio generated after the translation application translates the compressed audio; and carrying out sound mixing processing on the original sound to be translated and the translated audio to obtain a mixed audio, and outputting the mixed audio.

The present invention relates to a method for generating semantic adversarial samples based on GAN, comprising: S102. Pre-training a WGAN_GP neural network model using a local training set to obtain a generator and a discriminator; S104. Initializing noise data in a low-dimensional submanifold space; S106. Obtaining a generated image adversarial sample through the generator mapping; S108. Using the image adversarial sample to perform an adversarial attack on a target classifier including an attack term and a semantic term; S110. Determining whether the image classification output by the target classifier is the original classification category; if not, outputting a semantic adversarial sample, and if so, increasing the weight parameter of the attack term, and updating the initial value of the noise data, and iteratively repeating steps S106 to S110. The present invention searches for data in a low-dimensional submanifold space and performs semantic constraints in the image space by utilizing the approximation capability of GAN to data distribution, thereby ensuring the naturalness of the adversarial sample and generating an adversarial sample with high attack capability and retaining semantic information.

A method, system, storage medium and program product for speech enhancement in a mask, relating to the field of speech analysis or synthesis, the method comprising: collecting a first audio sample of a plurality of target objects in a mask-wearing state and a second audio sample in a mask-not-wearing state; determining the spectrum difference characteristics under the mask effect according to the first audio sample and the second audio sample; training a model based on the spectrum difference characteristics to obtain a speech enhancement model; obtaining a real-time audio signal of a target user in a work scene, and performing noise reduction preprocessing on the real-time audio signal to obtain a preprocessed audio signal; inputting the preprocessed audio signal into the speech enhancement model to obtain an enhanced audio signal; adjusting the enhanced audio signal according to a preset user configuration to obtain an output audio signal, and outputting the target speech as the output audio signal. Implementation of this application can improve the speech quality of the sound emitted in the mask and ensure smooth communication when multiple people work together.

The application discloses an image enhancement method and application thereof, the image enhancement method enhances the brightness constant processing on the collected image through a constructed camera response model, obtains an enhanced image, converts the enhanced image into a gray matrix image and then carries out denoising processing. The application also provides an edge extraction method based on the image enhancement method, uses a new SED model based on dynamic feature fusion to extract edges, that is, carries out semantic segmentation on the denoised gray image, converts the denoised gray image into a binary image, normalizes the amplitude scale of the multi-layer features of the binary image, carries out dynamic feature fusion, obtains the required edge features, and thus realizes edge extraction. The image enhancement method and the edge extraction method are both upgraded, and the application prospect is wide, and the image enhancement method and the edge extraction method can especially meet the demand of large batch and high precision product size detection, and reduce the artificial pressure.

The application discloses an active defense method, device and equipment for diffusion speech conversion and a medium, and relates to the technical field of speech security. The active defense method comprises the following steps: obtaining source speech and reference speech, introducing a constrained protection disturbance into the reference speech, constructing protected speech, and ensuring that the protection disturbance satisfies a disturbance imperceptibility constraint. The source speech and the protected speech are input into a speech conversion model with the reference speech as a speaker condition, content information analysis, speaker condition constraint acoustic generation and waveform synthesis are completed by the speech conversion model, and a protection generated speech is output. A joint loss function is constructed based on speaker embedding representations of the protection generated speech and the reference speech, gradients of a current disturbance position and a predicted midpoint position are calculated according to the joint loss function and are fused to obtain an update direction, the protection disturbance is iteratively optimized, and projection or clipping is performed to satisfy the disturbance imperceptibility constraint. Finally, the protected speech is output.

The invention provides a working face dynamic machine-following panoramic video stitching method and device. The method comprises the following steps: calling an original video stream of a target camera; wherein the target cameras are a plurality of cameras which are determined according to the position information of the coal mining machine and are within a preset range with the coal mining machine as the center; mapping the plurality of original video streams to a global coordinate system of the same working face according to the position information through a dynamic coordinate system conversion algorithm to obtain a machine-following panoramic video stream; performing local correction on the machine-following panoramic video stream through a local deformation correction algorithm to obtain a corrected video stream; fusing the overlapping areas of the adjacent cameras in the machine-following panoramic video stream through a multi-band fusion algorithm to obtain a fused video stream; the dynamic panoramic video which takes the coal mining machine as the center and continuously changes along with the movement of the coal mining machine is generated, and the remote control operation efficiency is remarkably improved. And the visual quality and naturalness of the dynamic panoramic video are ensured.

The invention provides a method for training the visual clue-dependent positioning reward capability of animals by using an animal behavior training and testing system, and the system realizes that one set of system covers multiple types of experiments through the opening and closing logic of an access door of an adjusting arm, the reward region reward putting rule and the free conversion of visual clues. A visual clue-reward association learning scene is constructed, and animals are guided to establish memory association between clues and reward positions. A three-stage visual clue navigation and memory ability testing method is established, the memory retention ability associated with clue rewards is evaluated in the association stage, the spatial memory ability is independently evaluated in the blank stage, and the regulation and control mechanism of visual clues on animal spatial memory is analyzed in the non-association stage.

The invention discloses an image enhancement processing method and system based on deep learning, and relates to the technical field of image enhancement, and the method comprises the following steps: obtaining a power equipment image, carrying out the coding and decoding of the power equipment image, and obtaining a plurality of edge domain and one core domain decoding image; performing feature extraction on the edge domain decoded image and the core domain decoded image through a preset feature extraction network; according to the image enhancement processing method based on deep learning, the problems of poor image quality and low resolution of the power equipment are solved, and the image quality of the power equipment is effectively enhanced.

The invention relates to the technical field of writing, in particular to a handwriting self-adaptive correction system for a tablet personal computer, which comprises a handwriting input module; the template determination module is used for determining a handwritten target template; the handwriting feature acquisition module is used for acquiring the pause time of the writing action; a handwriting correction module; a correction display module; the feedback module is used for determining the correction angle and the correction smoothness in the process of correcting the characters into the handwritten target template; and a receiving module. The handwriting input module and the template determination module are used for comparing the input characters with the font template to determine the correction template, the handwriting feature acquisition module is used for determining the correction moment, and the handwriting correction module and the correction display module cooperate to correct and display the handwriting. And finally, a feedback module and a receiving module feed back the correction angle and the correction smoothness to the user, and perform correction based on user adjustment and determine whether a new handwriting template is generated or not, so that correction accuracy is ensured.

The embodiment of the invention discloses a face image acquisition method and device, a storage medium and a camera. The method comprises the steps of collecting an initial image, performing face recognition on the initial image, and judging whether a face is recognized or not; when the face is recognized, counting brightness information and color information of the face based on the recognized face contour, calculating an exposure parameter combination according to the brightness information, and calculating a white balance gain according to the color information; and acquiring a face image based on the exposure parameter combination and the white balance gain. By applying the scheme provided by the embodiment of the invention, white balance adjustment and automatic exposure can be carried out only for the face area, and possible distortion caused by global adjustment is avoided, so that a face image meeting requirements can be acquired. And in addition, the naturalness of the face color can be kept under various different illumination and background colors, so that the method has strong adaptability.

The invention provides a DSA (Digital Subtraction Angiography) coronary artery stenosis analysis method based on matching degree measurement. According to the method, image preprocessing, catheter calibration, blood vessel segmentation and center line extraction are carried out on a DSA image, an improved MoM (Measure of Match) model is combined to accurately measure the diameter of a blood vessel based on the matching degree, and the stenosis rate of a stenosis area is calculated. The method has an automatic correction function, can automatically correct the narrow area according to the reference diameter of the normal blood vessel after the narrow detection, and improves the detection precision and stability. The method is suitable for rapid and accurate analysis of coronary artery stenosis, can effectively reduce manual operation errors, and is high in degree of automation.

The invention provides a fracture image auxiliary analysis software performance testing method and system, and relates to the technical field of software testing. The method comprises the following steps: acquiring fracture image test data; enhancing the fracture image test data to obtain fracture image enhanced data; determining the identification accuracy and the processing speed of the to-be-tested fracture image auxiliary analysis software on the annotation information; extracting regions of interest of each fracture image in the fracture image enhancement data in combination with the inertia moment matrix and the rotating rectangular frame to obtain a region of interest set; inputting the data into software for repeated testing to obtain secondary identification accuracy; under the condition that the secondary identification accuracy is greater than the preset secondary identification accuracy, determining that the performance of the to-be-tested fracture image auxiliary analysis software reaches the standard, otherwise, outputting an abnormal category corresponding to the annotation information; and determining that the performance of the to-be-tested fracture image auxiliary analysis software does not reach the standard, and outputting an abnormal category. And the test coverage range, the test speed and the test accuracy are improved.

The present invention relates to the field of geographic information security technology, and in particular to a vector map watermark embedding method and system. The method comprises: constructing a vector map learning model based on a deep neural network architecture including an encoder and a decoder, wherein the input of the model is an original vector map, the output of the encoder is the geographic features of the original vector map, and the output of the decoder is a vector map containing geographic feature data; training the vector map learning model; inputting the original vector map into the trained vector map learning model, outputting a vector map containing geographic feature data, recorded as a first vector map; generating a watermark bit stream using a hash algorithm; and embedding the watermark bit stream into the geographic feature data of the first vector map using a strategy of dynamically adjusting the watermark embedding strength, thereby obtaining a vector map after watermark embedding, which can adapt to the structural characteristics of the vector map and has high robustness and transparency of watermark embedding, thereby realizing safe and reliable vector map data protection.

This invention discloses a bit-flipping attack method and apparatus targeting large language models, relating to the field of large language model security technology. The method includes: generating an attack dataset containing questions using a large language model; inputting the attack dataset into the target large language model for forward propagation, outputting text data; constructing a perplexity loss function based on the text data; using a part-of-speech tagger to filter keyword elements in the text data, obtaining processed keyword elements; constructing a keyword element loss function based on the processed keyword elements; integrating the perplexity loss function and the keyword element loss function to obtain a total loss function; calculating the gradient value of each parameter based on the total loss function; and using a progressive bit search method based on the gradient values ​​to search for vulnerable bits in the target large language model to complete the bit-flipping attack. This invention can effectively reduce the accuracy of the output while maintaining the naturalness of the target large language model's output.

The application relates to the technical field of artificial intelligence, in particular to an intelligent audio noise reduction system based on an attention mechanism, which comprises an audio input end, a feature extraction module, a noise reduction module and an audio output end; the audio input end is used for acquiring a noisy audio signal and outputting the noisy audio signal; the feature extraction module is used for performing multi-scale time-frequency feature extraction and feature fusion on the noisy audio signal; the noise reduction module comprises a time sequence modeling unit, an attention unit and a noise estimation and suppression unit. The application can effectively distinguish speech and noise, avoid speech distortion, intelligently adjust attention weights, pertinently retain speech dominant area features and suppress noise dominant areas, iteratively converge to improve noise reduction effect and output high-quality noise reduction audio.

The invention relates to the technical field of image data processing, and discloses a target detection method, system and device based on an image data augmentation and image vision enhancement algorithm and a medium. A parameter prediction sub-module (Pe-PP module) is adopted to carry out visual enhancement preprocessing on the image on the basis of parameters provided by the parameter prediction sub-module (Pe-PP module), then a visual enhancement sub-module (VE module) is adopted to carry out contrast, brightness and white balance adjustment, and finally a target detection result is output; the target detection system comprises a target database construction module, a joint training module and a target detection module. The equipment and the medium are used for storing a computer program, and a target detection method is realized when the program is executed; according to the method, the problem of target category imbalance can be relieved, the overlapped target detection capability of the model and the overall detection precision are improved, and the method has good performance for solving the visual quality problems of image target category imbalance, low brightness and the like.

The embodiment of the invention provides a head-mounted display equipment control method and device, equipment and a storage medium, and belongs to the field of head-mounted display equipment. The method comprises the steps that under the condition that the head-mounted display device displays virtual content, in response to the change of the head posture of a wearer of the head-mounted display device, the current three-axis posture and the target anti-shake axial direction of the head-mounted display device are obtained, the target anti-shake axial direction is determined according to the current use scene of the head-mounted display device, and the target anti-shake axial direction is determined according to the current use scene of the head-mounted display device; the target anti-shake axial direction comprises at least one of a transverse rolling axis, a pitching axis and a yaw axis of the head-mounted display equipment; compensating the attitude of the target anti-shake axial direction in the current three-axis attitude to obtain a target three-axis attitude of the head-mounted display equipment; and according to the target three-axis attitude, controlling the head-mounted display device to adjust the pose and / or perspective deformation of the currently displayed virtual content. According to the technical scheme of the embodiment of the invention, the flexibility of attitude compensation of the head-mounted display equipment is improved.