4results about How to "Robust identification" patented technology

The application discloses a positive sample feature comparison method based on a twin neural network, and relates to the technical field of feature comparison.The method comprises the following steps: collecting positive sample data based on a target entity to be learned; performing strong and weak enhancement transformation on the positive sample data through a double-path asymmetric enhancement strategy, generating a strong enhancement view and a weak enhancement view for each positive sample data; performing feature extraction on the strong enhancement view and the weak enhancement view through a twin neural network, obtaining a query feature vector and a target feature vector; calculating a regularization contrast loss based on the query feature vector and the target feature vector through a regularization contrast loss algorithm; updating the twin neural network parameters through a back propagation algorithm based on the regularization contrast loss; and repeatedly iterating to obtain a feature extraction network.The application can realize more accurate and more robust similarity measurement and abnormality recognition, and improve the practicability and performance upper limit of the model under the condition of limited data.

The application discloses a cell microdroplet identification method based on XGBoost, and relates to the technical field of single-cell RNA sequencing, wherein step one is to search for the boundary of cells and empty droplets by using a RankMSE index and to construct a training set, and to solve the class imbalance problem in the training data by using multiple rounds of downsampling; step two is to combine the gene expression of cells and the cell quality control features calculated in advance such as cell entropy, and to use a machine learning method XGBoost to construct a cell-empty droplet binary classification model suitable for the current data; and step three is to iterate the cell-empty droplet binary classification model, add newly predicted empty droplet data to the model for retraining, and obtain an optimized prediction model. The cell microdroplet identification method based on XGBoost has the advantages that cells can be robustly identified in different data sets, and empty droplets and low-quality cells or cell fragments in the data can be effectively excluded, and the method has higher accuracy and stability compared with the prior art.

This invention relates to the field of bioinformatics and discloses a training method and apparatus for a deep learning-based analysis model of non-insulin-dependent diabetes mellitus (NIDM). The method involves acquiring target proteome data, which has binary labels characterizing the correlation between the target proteome data and NIDM; standardizing the features of the target proteome data; and performing data balancing using an oversampling algorithm to ensure that the number of target proteome data corresponding to the binary labels is equal. Based on the balanced target proteome data, the deep learning-based NIDM analysis model is trained to obtain the analysis model. By using the balanced, high-quality data to train the deep neural network model, the nonlinear correlations between high-dimensional protein features are fully explored, achieving accurate identification and analysis of the correlation between proteins and NIDM.

The application discloses a hyperspectral laser radar point cloud edge effect detection and filtering method, and relates to the technical field of laser radar remote sensing data processing.The application can robustly identify all points affected by edge effect at all wavelengths by constructing a multi-wavelength edge threshold automatic detection method based on an intensity distribution histogram.Further, by fine edge detection based on neighborhood convolution, points that are misdetected due to low reflectivity on the surface of leaves are effectively removed.Finally, by spherical space filtering, the intensity values of edge points are corrected by using the intensity information of the points inside the leaves around the edge points.The experimental results show that after the method is processed, the standard deviation of the intensity of the edge region of the leaves is reduced by 22.68%, the coefficient of variation is reduced by 28.30%, the coefficient of variation of all wavelengths is less than 1 (the average is 0.7288), and the consistency and quantitative reliability of the spectral data of the edge region are significantly improved.The method has a clear and simple operation process, can be embedded into an existing hyperspectral laser radar data processing process, and has important significance for improving the precision of three-dimensional inversion of vegetation biochemical parameters.