6 results about "Distance classifier" patented technology

The invention provides a fault diagnosis method and system for few-sample incremental equipment in an open set environment. The method comprises the following steps: pre-training a basic model comprising a feature extractor and a cosine similarity classifier based on a basic vibration sample; training a complementary model based on the enhanced sample and the pseudo-increment sample; the complementary model comprises a feature extractor added with a CBAM module and a square Euclidean distance classifier; inputting a basic vibration sample into the basic model and the complementary model at the same time, and fusing the dual-model output probability; a trained dual-model network is obtained through minimizing a loss function; inputting real incremental data into the trained dual-model network, freezing basic model parameters, and finely adjusting the last two convolutional layers and the classifier of the complementary model; and fusing the fine-tuned dual-model output probabilities to generate a final fault classification result. According to the method, intelligent fault diagnosis under the condition of continuously introducing a new type of data set can be realized, the applicable condition is more practical, the robustness is high, and the accuracy is high.

The application discloses a kind of variable working condition diagnosis methods of rotary reducer based on multi-view transfer, belong to fault diagnosis technical field.The method is first to the multiple visual vibration data collected is preprocessed and constructs variable working condition data set;Through fourier transform, extract each visual spectrum feature;Typical correlation analysis is used to construct embedding class discriminant transferable feature objective function, combined with maximum mean difference technique reduces the distribution difference between training and testing field, and introduces multi-view consistency constraint;Through generalized feature decomposition, solve common subspace projection, extract multi-view features with discriminant and transferability;Finally, nearest distance classifier is used to realize the fault diagnosis under variable working condition.The application effectively solves the problem that rotary reducer has poor generalization ability due to data distribution difference under variable working condition, improves the accuracy and reliability of fault diagnosis.

The invention relates to a non-ideal myoelectricity gesture recognition method based on Riemannian manifold transfer learning, which is characterized by comprising four parts: 1) constructing a covariance matrix by using spatial features of a myoelectricity acquisition channel as a basis; 2) constructing a Riemannian geometric manifold structure in a Riemannian space by taking the covariance matrix as a positive definite matrix; 3) constructing a Riemannian geometric center of the data and aligning the geometric center in a manifold space to realize transfer learning of the data; and 4) classifying different gestures by using a minimum Riemannian distance classifier combined with Fischer linear discriminant filtering. According to the method, a Riemannian manifold alignment transfer learning method is applied to gesture recognition of electromyographic signals, the influence of non-ideal factors such as electrode displacement and muscle fatigue existing in electromyographic signal recognition is solved by aligning manifold structures of a source domain and a target domain, and an effective electromyographic gesture recognition model is established.

The application provides a kind of cross-condition rolling bearing fault targeting migration diagnosis method and system, solve the problem that traditional rolling bearing fault diagnosis algorithm is difficult to extract network deep feature information in source domain and target domain, cannot realize effective cross-domain fault diagnosis.This application uses feature encoder to accurately extract high-dimensional mapping features of the signal from the input rolling bearing vibration signal;Further input the feature to the graph construction layer, mine the deep features of the data, and model the instance graph using the multi-channel kernel graph convolution network;Use the training based on difference and confrontation to minimize the distance between the source domain and target domain distribution, and the classifier uses the extracted domain invariant feature to complete cross-domain fault recognition.Compared with other methods, under the cross-condition of rolling bearing, the deep features can be better extracted for cross-domain transmission, greatly improving the diagnosis accuracy.

The application discloses a kind of based on graph regularization CNN-BiLSTM multi-sensor fault diagnosis method, medium and equipment, it is related to the computer fault diagnosis system field based on specific calculation model.The method comprises the following steps: collecting the condition monitoring data of multi-sensor, and the data collected is preprocessed, to obtain training set, verification set and test set;Establish CNN-BiLSTM network, graph regularization item is added in the nearest fully connected layer of distance classifier of CNN-BiLSTM network, complete GR-CNN-BiLSTM model construction;The data in training set is used to train GR-CNN-BiLSTM model, the data in verification set is used to evaluate GR-CNN-BiLSTM model, and the network parameter when the performance of GR-CNN-BiLSTM model is optimal is obtained;The data of test set is input into the GR-CNN-BiLSTM model of optimal performance and carries out fault diagnosis, to obtain fault diagnosis result.The application improves training efficiency and diagnostic accuracy, overcome the low training efficiency of existing deep graph regularization fault diagnosis method.

The application discloses a water body suspended matter spatial distribution monitoring method based on a dual-polarized SAR feature and belongs to the field of polarized synthetic aperture radar image processing. The application uses dual-polarized data to study the suspended silt target, aims to explore the role of the polarized data in identifying the suspended silt target. A preliminary classification map is obtained by using a new classification plane A' / alpha, A' has a more intuitive physical meaning compared with polarization entropy (H), then a likelihood ratio distance classifier capable of distinguishing weak scattering targets is used for further clustering to accurately lock the research range, and the polarized features (Shannon entropy intensity components) that are helpful to the identification of the suspended silt are analyzed, locked and used to accurately extract the suspended silt area.