9results about How to "Good noise robustness" patented technology

The invention provides a noise robustness streaming speech recognition method and model fused with a speech enhancement front end. The training method of the model comprises the following steps: acquiring a speech recognition training set; processing the speech training sample by using a causal speech enhancement network to obtain target filter bank features containing global information; processing the voice training sample by using a feature extraction method based on a filter bank to obtain original filter bank features; performing dynamic weighted fusion processing on the target filter bank features and the original filter bank features by using an attention mechanism network to obtain target fusion features containing supplementary information in the voice training sample; processing the target fusion feature by using an initial speech recognition network to obtain training text recognition information; generating a target loss result according to the target filter bank features, the training text identification information and the identification text label; and iteratively adjusting network parameters of the initial speech recognition model according to the target loss result to obtain a speech enhancement recognition model.

The invention provides a multi-type lost data recovery model construction method and system, and the method comprises the steps: obtaining data of a target sensor in a structural health monitoring (SHM) system, carrying out the preprocessing of the data of the target sensor, and obtaining a recovery data set; inputting the recovery data set into a pre-constructed data recovery model for training, and adjusting the weight of the pre-constructed data recovery model by using a loss function to obtain a final multi-type lost data recovery model; the method further comprises the steps of performing data recovery on data of a missing area in a structural health monitoring (SHM) system through the final multi-type lost data recovery model, and restoring the obtained data to obtain a final data result. According to the method, random point data missing, continuous fragment data missing and multi-channel and even full-channel data missing can be processed at the same time when the structural vibration data is recovered, and the method has relatively high comprehensive recovery capability and relatively high precision.

According to the method, beam structure crack positioning is carried out by using bilinear characteristics caused by a breathing effect of fatigue cracks, vibration response is decomposed into a closed component and an open component by using technologies such as zero point search and signal splitting and overturning, irregularity in split components is amplified by using wavelet transform, and the accuracy of the beam structure crack positioning is improved. A new crack damage index is constructed on the basis, a model reference of a healthy beam structure is not needed, vibration information in a healthy state does not need to be tested, interference of common steps, grooves and other factors on the beam structure can be avoided, and the method has the advantages of being good in noise robustness, rapid and reliable.

The invention discloses a fault diagnosis method based on acoustic signals of a carrier roller bearing of a belt conveyor, and the method comprises the steps: carrying out the self-adaptive global optimization of key parameters in a variational mode decomposition algorithm through employing a raccoon optimization algorithm, obtaining a global optimal solution, feeding back the global optimal solution to the variational mode decomposition algorithm, and obtaining the acoustic signals of the carrier roller bearing of the belt conveyor. Carrying out adaptive signal decomposition on the acquired carrier roller bearing acoustic signal by using the algorithm to obtain an IMF; screening components in the IMF by using the comprehensive index to obtain an optimal IMF component, and calculating a time domain feature value of the optimal IMF component to form a denoised signal feature vector; and optimizing hyper-parameters of the convolutional neural network by using a raccoon optimization algorithm to obtain an optimized convolutional neural network fault diagnosis model, and inputting the denoised signal feature vector into the model for fault classification diagnosis to obtain a diagnosis result. According to the method, the acoustic signal of the carrier roller bearing can be decomposed in a self-adaptive manner, the fault features are extracted, and the problems of mode aliasing and the like are effectively solved.

The present invention proposes a mining subsidence basin information extraction method and system based on InSAR, which relates to the field of mining area monitoring. The specific scheme includes: obtaining an InSAR interference map of the mining area to be monitored; inputting the InSAR interference map into a trained deep learning network model to extract mining subsidence basin information; wherein, the training data set of the deep learning network model includes a real training data set and a simulated training data set, wherein the real training data set is constructed by combining differential interferometry with a static probability integral method to obtain real mining subsidence basin information, and the simulated training data set is constructed by simulating the mining subsidence phase using a dynamic probability integral method and then combining Perlin noise and Gaussian noise models; the present invention fully trains the deep learning network model based on the real training data set and the simulated training data set to improve the accuracy and efficiency of mining subsidence basin information extraction.

This invention discloses a method and system for optimizing speech recognition under noise based on ultrasonic micro-motion detection. The method includes the following steps: an ultrasonic transmitting module continuously transmits ultrasonic signals; an ultrasonic receiving module continuously receives ultrasonic signals; the received ultrasonic signals are preprocessed to obtain the difference frequency signal of the reflected wave from the moving object; based on the obtained difference frequency signal, whether there is micro-motion within a specific distance is detected; if there is micro-motion, speech is output; otherwise, no speech is output. In this invention, the presence or absence of target speech is indicated by detecting lip micro-movements, which has good noise robustness. It utilizes the built-in speaker and microphone of a smart device to transmit and receive ultrasonic waves to detect lip micro-movements, requiring no additional cost, and has high accuracy and strong practicality, making it suitable for industrial-scale application.

The invention provides a human body internal hemorrhage position detection method and system based on an ultrasonic image, and relates to the technical field of ultrasonic detection.The human body internal hemorrhage position detection method based on the ultrasonic image comprises the following steps that an image data set is constructed through the collected ultrasonic image containing a bleeding point; based on the image data set, constructing and training an improved neighborhood attention converter model; acquiring a to-be-detected ultrasonic image, and performing bleeding position detection by using the trained improved neighborhood attention converter model to obtain a detection result; and carrying out visual display on the detection result by utilizing a visual tool. According to the invention, the deep learning network is applied to the hemorrhagic point detection of the ultrasonic image, and under the support of a data set, the deep learning network can obtain higher detection precision and better noise robustness compared with a traditional algorithm, and the network design is specially optimized for a small target detection task of the hemorrhagic point. And a better detection rate can be obtained.

The invention discloses a cross time-frequency skeleton separation method based on an optimal pixel connection rule. The method comprises the following steps: firstly, obtaining a complete skeleton of signal time-frequency distribution through mathematical morphology operation; secondly, capturing single-layer connected skeleton points in a non-cross region by using a skeleton tracking scheme based on eight pixel neighborhoods; and finally, a new weighted direction skeleton tracking strategy is utilized to determine an optimal skeleton communication direction, so that correct separation and extraction of a crossed region skeleton are realized, the problem that frequency modulation characteristics and frequency dispersion characteristics are difficult to extract at the same time by a traditional ridge extraction method can be effectively solved, accurate separation of a frequency modulation-frequency dispersion component crossed skeleton is realized, and the accuracy of the frequency modulation-frequency dispersion component crossed skeleton is improved. And the method has good noise robustness.

The present application belongs to the technical field of medical image processing and artificial intelligence, and particularly relates to a frequency domain enhanced in-situ biological hyperspectral feature extraction and classification method. By fusing the adaptive spatial-frequency feature extraction capability of fractional Fourier transform and the local-global spectral attention mechanism, the medical hyperspectral image can be deeply featured and efficiently classified, and the rapid and accurate diagnosis of early pathological lesions can be realized. The method is particularly suitable for real-time analysis of complex non-stationary pathological signals and multi-scale spectral-spatial information in a clinical environment, and provides accurate lesion classification and diagnosis assistance for doctors.