53 results about "Invariant feature extraction" patented technology

The invention discloses a speech recognition method based on a domain-invariant feature, and the method applies a speech domain-invariant feature extraction model to an end-to-end speech recognition model. The feature extraction model used in the speech recognition method aims at the robustness problem, and by adding more types of speech data to train the speech feature extraction model, better parameters can be obtained and a better domain-invariant feature extraction model can be obtained. The speech recognition method based on the domain-invariant feature uses unlabeled pure speech data totrain the feature extraction model, uses a small number of speeches with text annotation to train an end-to-end acoustic model, and provides important technical support for improving the robustness ofthe end-to-end acoustic model. Compared with the prior art, the speech recognition method based on the domain-invariant feature has higher recognition accuracy in different noise environments, smaller task quantity of speech annotation tasks, and faster training and testing speed of the models.

The invention provides a real-time target matching tracking method and a system. According to the method, feature information of pixels contained in each of at least two images which are taken in the same place at similar shooting time is analyzed, so that at least one target area from each image is acquired; and then matching target areas of each target area in the corresponding image according to area information occupied by the target area in the corresponding image are determined, wherein the matching target areas are positioned in other images. The areas, which is occupied by a target, in the images are used as tracking objects, the matching target areas, which are positioned in the other images, of each target area are determined based on the area information occupied by the target area in the corresponding image, existing invariant feature extraction algorithm is omitted, accordingly, compared with the prior art, the method is low in complexity, real-time tracking to the target areas can be realized easily, and the method can be applied to more monitoring places.

The invention discloses an open set domain adaptive image classification method and system in the technical field of image recognition, and the method and system enable a network to better obtain domain invariant features through a channel attention mechanism, facilitate the migration of the features, and enable the features to be trained more easily. The method comprises the following steps: respectively inputting a labeled sample obtained from a source domain and an unlabeled sample obtained from a target domain into a feature extractor based on a channel attention module to obtain a weighted multi-channel feature map; sending the weighted multi-channel feature map into a label classifier, dividing samples with labels into K known categories, and dividing samples without labels into K known categories visible in a source domain and an unknown category invisible in the source domain; sending known categories from a source domain and a target domain to a domain discriminator, and enhancing domain invariant feature extraction based on the generative adversarial network; reducing inter-domain differences between the source domain and the target domain based on covariance matching.

The invention discloses a frequency displacement invariant feature extracting method considering multifactor in voice, and the method is used for distinguishing text-independent speakers under a complex environment. The method comprises the step of: in consideration of the time, frequency, scale and phase information of voice, performing multifactor characterization on a voice signal energy spectrum through two-dimensional plurality wavelet transform of different scales and phases, in consideration of the displacement invariant feature of the frequency, calculating a displacement invariant feature projection matrix on a frequency order through a convolution-type non-negative tensor decomposition method to obtain a multifactor sparse feature, decorrelating the feature through discrete cosine transform, and calculating the first order and second order difference coefficients of the feature, thus finally obtaining speaker feature with good robustness. The frequency displacement invariant feature extracting method considering multifactor in voice disclosed by the invention calculates the multifactor frequency displacement invariant feature with robustness through the convolution-type non-negative tensor decomposition method to distinguish the text-independent speakers under a noise environment, so that good distinguishing accuracy is good.