7 results about "Quaternion matrix" patented technology

The present disclosure provides a method for determining the pose of an optical master hand, a surgical robot system. The determination method comprises: acquiring current quaternion data of the optical master hand at the current time; based on the current quaternion data and historical quaternion data of the optical master hand at a plurality of historical time points before the current time, determining a target quaternion data set; based on the target quaternion data set, determining a first quaternion matrix, the first quaternion matrix comprising the target quaternion data set, a target velocity set of the target quaternion data set, and a target acceleration set of the target quaternion data set; embedding the first quaternion matrix into a high-dimensional space to obtain a second quaternion matrix; performing position encoding on the second quaternion matrix to obtain a third quaternion matrix; based on the third quaternion matrix, using a network model to determine a target quaternion matrix with a dependency relationship; based on the target quaternion matrix, obtaining a target Euler angle, the target Euler angle representing the pose of the optical master hand at the current time.

The invention relates to the technical field of image fusion processing, and discloses a multi-source fundus lesion image characteristic-based image collaborative fusion processing method, which comprises the following steps of: performing scale normalization processing on a color fundus image and tomography data through resampling to obtain a standard color fundus image and standard tomography data; the method comprises the following steps: respectively fitting an inner boundary membrane curved surface for standard tomography data, expanding to a plane, performing maximum intensity projection according to a preset layer thickness range, and generating a two-dimensional structure layer image; extracting a normalized gray matrix of the standard color fundus image, combining the normalized gray matrix with the normalized two-dimensional structure layer image, and constructing a quaternion matrix containing color and structure information; performing pyramid decomposition on the quaternion matrix to obtain a pyramid high-frequency layer and a pyramid low-frequency layer; and determining the focus energy weight based on the gradient significance of the standard color fundus image and the standard tomography data and the local energy of the high-frequency layer and the low-frequency layer of the pyramid.

The application discloses a quaternion neural network color image restoration method based on low rank and gradient smoothing, comprising the following steps: converting a color image to be processed into a pure virtual quaternion matrix; constructing a damaged color image data and a data set of a pixel loss and a noise scene; constructing a symmetric structure quaternion auto-encoding network model, wherein an encoder is composed of four QCNN layers, a decoder is composed of four QDCNN layers, an activation function layer and a bias term are connected between each QCNN layer of the encoder, the decoder reconstructs an image through reverse rotation transformation, a quaternion convolution kernel is subjected to rotation-scaling parameterization and is initialized by using an adaptive normal distribution; constructing a quaternion image restoration optimization model fusing a low rank priori and a smoothing priori, introducing a quaternion gradient kernel norm regularization term; constructing a completion model and a denoising model based on the optimization model, and solving the models by using ADMM iteration; and mapping an iteration result into a repaired RGB color image and outputting the image. The application can realize high-quality restoration of a color image missing or contaminated by noise.

The application provides a model reasoning method and device, electronic equipment and storage medium, and relates to the technical field of artificial intelligence, comprising: converting lossless equivalence of real number matrix multiplication in a pre-trained real number model into quaternion matrix multiplication to obtain a quaternion model; quantizing quaternion weights of the quaternion model; performing quantization-aware training fine-tuning on the quantized quaternion model; and performing multiplication-free reasoning based on the quantization-aware training fine-tuned quaternion model. The application can achieve extremely high model compression rate while maintaining or even improving model accuracy, significantly accelerating model reasoning speed, and improving model reasoning performance.

The application discloses a kind of based on quaternion convolutional neural network's no reference image quality evaluation method and system, method includes: based on the improved ResNet network of preestablished to at least one pure quaternion matrix is extracted, and the extracted feature is fused, obtains second target scale feature, again with second target scale feature and the third scale feature is secondly fused, obtains third target scale feature, respectively in the feature vector in the first scale feature, second target scale feature, third target scale feature and fourth scale feature after extraction processing are handled, and each feature vector after extraction is aggregated, obtains target feature vector, target feature vector is input into the preestablished fully connected neural network, and the quality score corresponding to target feature vector is mapped and output by preestablished fully connected neural network.It improves the accuracy and robustness of no reference image real distortion.

The application provides an image feature extraction and privacy protection identification method, comprising the following steps: encoding different components of a privacy image into a quaternion matrix; performing multi-resolution singular value decomposition in the form of quaternions on the quaternion matrix to obtain a first approximate component, and performing multi-resolution singular value decomposition in the form of quaternions on the first approximate component to obtain a second approximate component; performing sparse random projection on the quaternion matrix, the first approximate component and the second approximate component; performing feature extraction on the results of sparse random projection through a concatenated quaternion two-dimensional discrete cosine transform network; and inputting the extracted features into a classifier for identification. Through the method provided by the application, multi-view or multi-modal image components can be effectively fused and the revocability can be met, the security is ensured, the reliable identification precision is ensured, and the method can be applied to the fields of multimedia information security and visual content protection.

The application discloses an image recognition method based on quaternion generalized kernel sparse principal component analysis, and the method comprises the following steps: acquiring training and test sample images; extracting the entropy, red, green and blue four component information of each image, and performing quaternion matrix representation on the information to construct a corresponding quaternion real representation matrix; constructing a corresponding quaternion covariance kernel matrix and a quaternion p-norm Euclidean distance according to the quaternion real representation matrix, and then constructing a quaternion generalized kernel sparse principal component analysis optimization model; solving the optimization model, taking the calculated kernel sparse principal components of the training sample in the row and column directions as the final solution; calculating the projection matrix of the training and test sample covariance kernel matrix according to the final solution of the model in the row and column directions; and using the quaternion p-norm Euclidean distance to recognize the category to which the images in the test sample set belong, so that the recognition accuracy and robustness are improved.