974results about How to "Improve denoising effect" patented technology

The invention discloses a cascaded residual error neural network-based image denoising method. The method comprises the following steps of building a cascaded residual error neural network model, wherein the cascaded residual error neural network model is formed by connecting a plurality of residual error units in series, and each residual error unit comprises a plurality of convolutional layers, active layers after the convolutional layers and unit jump connection units; selecting a training set, and setting training parameters of the cascaded residual error neural network model; training the cascaded residual error neural network model by taking a minimized loss function as a target according to the cascaded residual error neural network model and the training parameters of the cascaded residual error neural network model to form an image denoising neural network model; and inputting a to-be-processed image to the image denoising neural network model, and outputting a denoised image. According to the cascaded residual error neural network-based image denoising method disclosed by the invention, the learning ability of the neural network is greatly enhanced, accurate mapping from noisy images to clean images is established, and real-time denoising can be realized.

The invention discloses an ReLU convolutional neutral network-based image denoising method. The method comprises the following steps of building an ReLU convolutional neutral network model, wherein the ReLU convolutional neutral network model comprises a plurality of convolutional layers and active layers after the convolutional layers, wherein the active layers are ReLU functions; selecting a training set, and setting training parameters of the ReLU convolutional neutral network model; training the ReLU convolutional neutral network model by taking a minimized loss function as a target according to the ReLU convolutional neutral network model and the training parameters of the ReLU convolutional neutral network model to form an image denoising neural network model; and inputting a to-be-processed image to the image denoising neural network model, and outputting a denoised image. According to the ReLU convolutional neutral network-based image denoising method disclosed by the invention, the learning ability of the neural network is greatly enhanced, accurate mapping from noisy images to clean images is established, and real-time denoising can be realized.

The invention discloses a speech enhancement method for speech recognition in a noise environment, and the method comprises the steps: combining the improved MVDR beam forming which is based on the time-frequency masking and employs the speech time-frequency domain sparsity principle with the improved Wiener filtering, acquiring a speech signal of a microphone array, and constructing an MVDR beamformer based on time-frequency masking, making full use of the spatial information of the speech signal, enhancing the speech signal in a target direction, suppressing the interference of noise in other directions, and then removing the residual noise and improving the speech intelligibility through a modified Wiener filter. The method is applied to a voice recognition front end, and can effectively remove noise and improve the voice intelligibility, thereby improving the recognition rate of a speech recognition system, solving a problem of how to reduce the speech distortion in the noise environment and improve the speech recognition rate of the noise environment. The method can be applied to a household robot and intelligent voice equipment.

The invention discloses a convolutional neural network medical CT image denoising method based on residual error learning. The method comprises the following specific steps: 1) constructing a medicalCT image model; 2) constructing a neural network model; 3) training the network; 4) updating parameters; and 5) denoising the medical CT image: inputting the medical CT image containing the noise intothe constructed network model, and outputting the medical CT image without the noise through the network. The method has the advantages that medical CT image denoising is carried out by combining knowledge in the aspect of the convolutional neural network in deep learning. Noise in the image is approximately learned in a residual error learning mode, good pertinence is achieved, and meanwhile thetraining efficiency of the neural network is improved. The convolutional neural network and the residual error learning method are adopted, so that the feature information in the image can be betterlearned, and more local image information can be reserved in the image denoising process. And meanwhile, the image denoising capability is also improved.

This invention discloses an self-adaptive mike array system and a method for processing its phone signals, which adds an self-adaptive filter coefficient refresh control module in the mike system to control if the filter should refresh the coefficient or the step-size value when refreshing the coefficient, so that the refreshment of the coefficient is only against the noise so as to avoid the side effect of reducing effective phone signal amplitude while conducting filter coefficient refresh using existing technology when the noise component from the back signals is comparable with or smaller so than the phone component from frontage. This invented system applies a high-order adaptive filter to reduce the noise amplitude to the utmost degree and ensures the best noise-elimination effect no matter what positions the system zero-point directs when multiple noise sources exist.

The invention discloses a natural image denoising method based on dictionary learning and block matching, which mainly solves the problems that texture details are easily lost and homogenous areas are not smooth in the conventional natural image denoising. The method comprises the following steps of: (1) setting a denoising target function and inputting a noise-containing image z(x); (2) making an original image equal to the noise-containing image, namely y(x)=z(x), and making a dictionary D be a redundant discrete cosine transform (DCT) dictionary; (3) updating the atoms of the dictionary D and a corresponding coefficient matrix alphaij by using a kernel-singular value decomposition (KSVD) algorithm; (4) denoising the noise-containing image z(x) by using a block matching three-dimensional (BM3D) algorithm to acquire a primary denoising result; and (5) introducing the updated D and alphaij into the estimation formula of the original image to acquire the denoising result of the noise-containing image z(x). Compared with the conventional classic denoising method, the method achieves a better denoising effect and can be used for denoising a natural image; and the homogeneous area is smoothened, and the texture, the profile and the edge detail information of the image can be maintained at the same time.

The invention discloses an image denoising method and system based on deep learning. The image denoising method based on deep learning includes the steps: constructing a main neural network structureand an auxiliary neural network structure, respectively assigning the trainable parameter initial value of the first convolutional layer and the trainable parameter initial value of the fifth convolutional layer in the auxiliary neural network structure to the trainable parameter initial value of the first convolutional layer and the trainable parameter initial value of the 15th convolutional layer in the main neural network structure; adding a training set noise adding image into the main neural network structure after assignment, and obtaining a noise characteristic image by performing imagecharacteristic extraction, training and learning on the input training set noise adding image through a forward propagation algorithm; according to the noise characteristic image, determining a training model; inputting a verification set noise adding image into the training model, and outputting a final training denoising model; and adding a test set noise adding image into the final training denoising model to test, and obtaining a denoised image, thus greatly improving the denoising efficiency and the denoising effect.

The invention discloses a detecting method of a weak periodic signal based on a chaotic system and wavelet threshold denoising, which comprises the following steps of: firstly carrying out wavelet decomposition on collected information, and determining a decomposition scale according to an actual signal-to-noise condition; denoising a wavelet high-frequency coefficient after the wavelet decomposition, wherein in the process of wavelet threshold denoising, the selection of a threshold is an important problem and directly influences a denoising result, so that the invention firstly provides a method for determining the threshold according to the scale for carrying out coefficient threshold processing to improve the denoising effect; reconstructing a signal after denoising, merging the signal to be detected after wavelet denoising reconstruction as one part of the driving force of the chaotic system into a chaotic detecting system, further inhabiting noise interference by utilizing the characteristics of the chaotic system for strong noise immunity, periodic weak signal sensitivity and the like, and effectively extracting the weak signal. The invention improves the detection threshold and the signal-to-noise ratio purely based on the chaotic detecting system.

The invention relates to a partial discharge signal denoising method based on lifting wavelet transform, which includes the following steps: (1) a partial discharge signal to be denoised is inputted; (2) lifting wavelet decomposition is carried out on the partial discharge signal, so that high-frequency coefficient components of different decomposition scales and a low-frequency coefficient component of the highest scale are obtained; (3) wavelet entropy-based layered thresholds and a soft threshold function are adopted to quantify the high-frequency coefficient components in order to remove noise components, and the high-frequency coefficient components are stored as new high-frequency coefficient components; (4) the new high-frequency coefficient components and the low-frequency coefficient component of the highest scale obtained in step (3) are utilized to compose a coefficient component for signal reconstruction, signal reconstruction is carried out on the coefficient, and thereby a denoised partial discharge signal is obtained. Lifting wavelets are completely transformed in a time (space) domain, and high-pass and low-pass filters are turned into a series of relatively simple prediction and update steps. Therefore the denoising speed of lifting wavelet transform is high, the design is flexible and simple, and the partial discharge signal denoising method is easy to put into practice.

The invention discloses a far-field speech recognition processing method and device. The method comprises the steps that a far-field speech is received; the far-field speech is input into a neural network-based speech training model obtained through pre-training; interference removing processing is conducted on audio features of the received far-field speech through audio features of far-field speeches and near-field speeches in the speech training model, and the processed far-field speech is obtained; the processed far-field speech is recognized. According to the method, optimization processing on the far-field speech can be achieved, a better processing result can be acquired, and equipment cost input is reduced.

Embodiments of the invention disclose a speech de-noising method and a speech de-noising device. The method comprises the following steps: detecting the speech of a speech signal with noise to distinguish between speech frames and non-speech frames; estimating the noise of the speech frames and the noise of the non-speech frames to get a noise power spectrum fused estimated value, wherein the noise power spectrum fusion estimated value is the fused value of the noise power spectrum estimated value of the speech frames and the noise power spectrum estimated value of the non-speech frames; and de-noising the speech signal with noise according to the noise power spectrum fusion estimated value. According to the technical scheme provided by the embodiments of the invention, the noise of the speech frames and the noise of the non-speech frames are estimated, and the speech signal with noise is de-noised based on the noise estimation results of the speech frames and the non-speech frames. Thus, the de-noising effect of the existing speech de-noising scheme is improved effectively, and the quality of speech is improved.

The invention discloses an image de-noising method based on a compression convolutional neural network, including the following steps: constructing a training data set; constructing a compression de-noising convolutional neural network model; using the training data set to train the network model; and inputting a noisy image to the trained network, and subtracting the output image of the network from the noisy image to get a clear de-noised image. The de-noising convolutional neural network in the invention is mainly characterized in that the convolution layer of the original de-noising convolutional neural network is replaced with a convolution layer decomposed and compressed by a low-rank matrix. By improving the existing de-noising convolutional neural network DnCNN, the network parameters are reduced by at least 75%, the network is simplified, and an excellent de-noising effect is maintained.

The invention provides a real image blind denoising method based on a deep residual network. According to the method, an RGB spatial clear image set is selected through an image dataset, and an RGB spatial image group set is constructed through spatial transformation; images under multiple scenes are shot through multiple cameras, and real image groups are constructed according to real clear images and real noisy images shot by each camera under each scene, and a real image group set is constructed; multiple RGB spatial image groups in the RGB spatial image group set and multiple real image groups in the real image group set are randomly selected to construct an image training set, and a preprocessed image training set is obtained through preprocessing; remaining RGB spatial image groups in the RGB spatial image group set and remaining real image groups in the real image group set are used to construct an image test set; and the preprocessed image training set is used as input to construct an image denoising residual convolutional neural network, the neural network is trained in combination with residual learning and a batch normalization strategy, and the image test set is denoised. The method has the advantages that convergence speed is high, and the denoising effect is good.

The present invention relates to a denoising method of a synthesizing aperture radar image based on an independent component analysis basis image in the image processing technical field. First, the original image is sampled and an initial matrix is gained, and the equalizing value removal and whiten operation is performed, then the processing result is disposed as the input matrix of the single component analysis method, and the single component analysis method is performed, and a basis vector aggregation of the original image and the single component are gained; then the basis vector is converted into a relative basis image, and disposed in two steps by adopting the basis image as the object: step 1, the multifractal Herder exponent is adopted as the cost function to make the smooth enhance to the basis image; step 2, according to the signal separation concept, the separation criterion is proposed, the enhanced basis image is separated, and the basis image aggregation relative to the non noise signal is separated. The non noise basis image and the relative encoder matrix are reconstructed to gain the final denoising image. The present invention realizes the compromise between the SAR image noise removal and the useful information reservation, and gains good denoising performance.

The invention discloses a radar image based moving target morphology detecting and tracking method, which is characterized by firstly dividing gray scales after collecting one frame data, adopting a frame differential method to remove most fixed clutter noises, then adopting a sliding window morphology matching algorithm to isolate the target, accurately positioning the target through target centralization, producing binary images, then combining multiframe binary images with the movement law of the target to carry out track correlation, further removing the fixed target and noise interference and adopting second matching and forecasting to solve the problems of weak signal searching and noise interference, thus realizing lossless tracking. The method features simpleness, less calculated amount, obvious de-noising effect, timely target capturing and high success rate, and can realize real-time detection and lossless tracking of the moving target.

The invention belongs to the technical field of digital signal processing, such as seismic data processing, and particularly relates to a mall-scale threshold denoising method based on wavelet transform; and the theoretical basis is provided for selecting frequency dividing parameters and a proper wavelet threshold during wavelet transform denoising. The method includes firstly scanning seismic data in a small-scale manner, and acquiring a relevant coefficient in a time window; secondly setting a threshold, judging whether the data in the small-sized time window are mainly seismic signals or noise signals, and then adopting proper frequency dividing parameters and a wavelet threshold to denoise; when the seismic channels are mainly of the seismic signals, adopting conventional wavelet decomposition and a conventional hard threshold or soft threshold; when the seismic channels are mainly of the noise signals, performing wavelet packet decomposition on the seismic signals, and setting a following floating threshold method by adopting the best entropy principle; finally performing wavelet reconstruction on denoised wavelet scales, and acquiring a seismic channel set with higher noise ratio after denoising.

The invention discloses a method and device for performing adaptive denoising on a video image. The device comprises a video image reading module, an edge detection module, a filtering mode selection module and a filtering module; the method comprises the following steps of: reading in a frame of video image from a buffer zone, respectively calculating motion vectors between each pixel in the frame of image and the pixel at the same location in a neighboring frame, and respectively executing the edge detection on each pixel in the frame of image; respectively selecting the filtering mode having corresponding intensity for each pixel according to the edge detection result and the calculation result of the motion vector and of each pixel, and then, correspondingly filtering each pixel by a Gaussian similarity filter. The method and the device have obvious denoising effect on various noises of different scene sequences, and can effectively remove noises by combination of a time domain and a space domain.

The invention relates to the technical field of image processing, more specifically, to a depth convolution neural network image denoising method based on an Inception model. Building a training network with one-layer Inception model; setting the parameters of network solution and choosing the appropriate supervisory framework; the parameters of the network structure are optimized, and the obtained network is used for denoising. The depth convolution neural network image denoising method based on the Inception model of the invention is based on the depth convolution neural network and combinesGaussian white noise to simulate unknown real noise and the network containing the Inception model to reduce the parameter quantity on the basis of maintaining the original denoising effect; throughthe way of data expansion, the content of training data is enriched, and the neural network can learn the inner structure of image distribution or noise. By introducing the residual structure, the network converges more easily and the effect is the best.

The invention discloses a wavelet speech denoising method based on a new threshold function. The method mainly comprises steps: 1) noisy speech is generated; 2) wavelet decomposition is carried out on the noisy speech; 3) wavelet threshold processing is carried out on the noisy speech; 4) wavelet reconstruction is carried out on speech after processing; and 5) speech signals after denoising are obtained. The method focuses on processing in the step 3), and the step 3) directly determines denoising effects. In the step 3), a wavelet coefficient after wavelet decomposition on the noisy speech is subjected to threshold processing, which mainly involves threshold determination and threshold function selection. As the new threshold function is adopted, the problems that signals after processing generate oscillation as the traditional hard threshold function is discontinuous and great distortion happens after signals are processed by a soft threshold function can be solved. The new threshold function can better suppress noise, the signal-to-noise ratio of speech signals is improved, speech distortion is reduced, and better denoising effects are obtained.

The invention discloses a sparse representation and low-rank double restraints-based nonlocal denoising method. The problem of a poor effect in an existing denoising method is mainly solved. The method comprises the following implementation steps: (1) inputting a noise image; (2) estimating a noise standard deviation; (3) setting parameters; (4) obtaining a pixel block sample set; (5) building a similar block matrix; (6) obtaining a coefficient matrix of the similar block matrix; (7) carrying out singular value filtering on the coefficient matrix; (8) obtaining the denoised coefficient matrix; (9) obtaining the denoised image matrix; (10) judging whether the maximal iterative times is achieved, if so, carrying out a step (11), or else, carrying out residuals recovery, and carrying out the step (2); and (11) outputting the denoised image matrix. The method has mating constraints on the sparsity and the constitutive property of an image signal. Compared with the prior art, the method disclosed by the invention has the advantages that structural texture information of a natural image can be kept and recovered when the noise is well smoothed.

The invention relates to a method and a system for extracting field image edges of plant leaves. The method comprises the following steps of: 1, smoothing and denoising a subimage in a partition area adjacent to an area containing a single leaf serving as a target image; 2, performing area-based threshold partition and gradient-based edge detection partition on the image processed by the step 1, and performing initial optimization to acquire an inner template of a target area; 3, judging whether the target area of the image processed by the step 2 is a leaf image or not, if so, executing the step 4, and otherwise, performing grayscale mapping and morphological reconstruction on the image processed by the step 1, and executing the step 2; and 4, performing operation of adhesion disconnection, false edge removal and closure on the image processed by the step 2 to obtain the edge image with a closed edge. The method and the system improve the accuracy and integrality of the extraction ofthe field image edges, which have non-uniform grayscale, of the plant leaves.

The invention discloses an image de-noising method based on a sparse self-adapted dictionary, and the method is mainly used for overcoming the defects that over-fitting exists when an existing method is used for training the dictionary, and the self-adaption is insufficient. The realization process comprises the following steps of: (1) obtaining image blocks from images with noises and paralleling the image blocks into vectors to form a training data set; (2) utilizing the training data set to iteratively train the dictionary; in an iteration process, taking the dictionary obtained by iteration as a basic dictionary of the iteration, and after the iteration is finished, obtaining a final dictionary and an encoding coefficient matrix of the training data set on the dictionary; (3) utilizing the dictionary and the encoding coefficient matrix obtained by training to obtain a de-noised data set; and (4) utilizing the de-noised data set to reconstruct a de-noised image. The dictionary trained by the method disclosed by the invention has sparseness and better self-adaptation; the effect of de-noising the image is improved; and the method can be used for de-noising a natural image and a medical CT (Computed Tomography) image.