31results about How to "Reduce redundant features" patented technology

The invention discloses a fused attention model-based Chinese text classification method. The method comprises the following steps of: respectively segmenting a text into a corresponding word set anda corresponding character set through word segmentation preprocessing and character segmentation preprocessing, and training a word vector and a character vector corresponding to the text by adoptionof a feature embedding method according to the obtained word set and character set; respectively carrying out semantic encoding on the word vector and the character vector by taking a bidirectional gate circulation unit neural network as an encoder, and obtaining a word attention vector and a character attention vector in the text by adoption of a word vector attention mechanism and a character vector attention mechanism; obtaining a fused attention vector; and predicting a category of the text through a softmax classifier. The method is capable of solving the problem that more redundant features exist in the classification process as existing Chinese text classification methods neglects character feature information of texts, the extracted texts are single in features, all the pieces of semantic information of the texts are difficult to cover and features having obvious contribution to the classification are not focused.

The invention discloses a wireless sensor high-dimensional data anomaly detection method based on deep learning. The method comprises the following steps: (101) obtaining historical data; (102) establishing a DBN-QSSVM hybrid model; (103) training the mixed model by using historical data of the sensor; (104) collecting test data of a sensor to be detected; (105) using a trained DBN-QSSVM hybrid model in (103) to perform abnormality detection on sensor test data; (106) outputting abnormal data in sensor test data and (106) outputting abnormal data in the sensor test data. Relevant algorithms and processes in the prior art are improved. The method for achieving online detection technology when the high-dimensional data is processed is provided, the space complexity and time complexity can begreatly reduced under the condition that the accuracy of the data exception detection method is not reduced, and therefore the method is more suitable for large-scale high-dimensional data exceptiondetection.

The invention belongs to the field of computer vision and specifically relates to a space-time bone characteristic and depth belief network-based human body behavior identification method. According to the method, a bone sequence is obtained via a depth camera based on each motion; bone point time sequence characteristics are extracted from the bone sequence and comprise displacement characteristics, speed characteristics and acceleration characteristics of each bone point; a bone reference point is chosen, relative distance between each bone point and the bone reference point is calculated, space characteristics of a motion bone sequence is obtained, the bone point time sequence characteristics and the space characteristics of a motion bone sequence are subjected to clustering operation,a characteristic length is fixed, a global motion description operator is established and subjected to dimensionality reducing operation, and motion classification is finished based on cross validation and a classifier.

The invention discloses an indoor scene semantic segmentation method based on an improved full convolutional neural network. The method comprises the steps of firstly constructing a convolutional neural network, wherein a hidden layer of the convolutional neural network comprises five neural network blocks, five feature re-extraction convolutional layer blocks, five partitioning attention convolutional blocks, twelve fusion layers and four up-sampling layers; inputting the original indoor scene image into a convolutional neural network for training to obtain a corresponding semantic segmentation prediction graph; calculating a loss function value between a set formed by semantic segmentation prediction images corresponding to original indoor scene images and a set formed by one-hot coded images processed by corresponding real semantic segmentation images to obtain an optimal weight vector and an offset term of a convolutional neural network classification training model; and inputtingan indoor scene image to be subjected to semantic segmentation into the trained convolutional neural network classification training model to obtain a predicted semantic segmentation image. The methodhas the advantage that the semantic segmentation efficiency and accuracy of the indoor scene image are improved.

The invention discloses a social network text sentiment analysis method based on a deformable self-attention mechanism, which is used for analyzing the sentiment of a user utterance. The method comprises the following steps: segmenting each sentence in user utterance text data into words, and expressing each word with a word vector; inputting the word vector sequence into a bidirectional recurrentneural network (Bi-LSTM) to obtain a coded representation of each word; utilizing a deformable self-attention mechanism to convert the coding representation of the words into a plurality of sentencecoding representations with different context ranges; fusing the plurality of sentence code representations to obtain a sentence code representation; inputting the fused sentence code representation into a feed-forward neural network (FFN) for classification, and outputting a result; according to a model output result and a data real result, iteratively training a model updating parameter by minimizing a cross entropy loss function; and inputting the social network text to be classified into the trained model to obtain an emotion analysis result.

The invention discloses an escape vehicle rapid identification and tracking method based on a vehicle-mounted unmanned aerial vehicle. The method specifically comprises the steps of (1) constructing a prediction model for the position of a road section where an escape vehicle is located, and taking the first three areas with the largest probability values as candidate areas; (2) scanning vehicles in the three candidate areas and uploading the vehicles to a cloud end through a 5G network; (3) establishing a lightweight escape vehicle re-identification model based on the improved DenseNet121, and performing vehicle feature extraction and re-identification on a picture of each candidate area based on the escape vehicle to be queried; and (4) transmitting the position information of the candidate area where the picture with the highest escape vehicle re-identification rate is located to an on-duty person, and completing tracking of the hit-and-run vehicle. The unmanned aerial vehicle can rapidly and reliably re-identify the escape vehicle, and the case solving efficiency is improved.

The invention designs a traditional Chinese medicine image classification and retrieval method based on a pyramid network, and is used for solving various problems that traditional neural network classification focuses on global semantic information such as contours and shapes, and a lot of information is lost due to multiple down-sampling operations. In order to achieve an ideal classification effect, multi-scale feature fusion is realized through a pyramid structure of a feature pyramid, and the requirement that shallow-layer and deep-layer spatial and semantic information needs to be fused in traditional Chinese medicine decoction piece image classification is met. Aiming at the problems that a multi-scale feature fusion mode can cause a large number of redundant features and neural network training can be interfered by redundant information, an SE-Net model is used, the redundant features are reduced, a sampling strategy and triple loss are combined, a multi-task learning framework is used for classifying and retrieving images at the same time, and more compact and accurate traditional Chinese medicine image feature information can be obtained.

The invention discloses a crowd counting and positioning method based on a coding and decoding structure, relates to the field of computer vision, and solves the problems that in the prior art, characteristics are not fully utilized, and a tag graph cannot well give consideration to counting and positioning tasks. A multi-scale feature fusion module is introduced into the deep layer of the network, a space-channel attention up-sampling module is introduced into a re-decoding part, and the multi-scale feature fusion module captures features of multiple scales by using cavity convolution with different expansion rates and performs feature fusion, so that the robustness of the network to deal with scale change is improved; the space-channel attention up-sampling module guides superficial layer features to perform high-efficiency fusion through high-level high-level semantics, so that interference of redundant features and picture backgrounds is reduced; secondly, a new label graph is provided, and the label graph has the advantage of simple counting of a density graph and also has the positioning performance of an FIDT graph.

The invention discloses a support vector machine electrical network intrusion detection system and method based on feature selection. The system comprises an intrusion information database, a data preprocessing module and a data analysis module, wherein the intrusion information database, the data preprocessing module and the data analysis module are connected through network equipment; the intrusion information database is used for storing intrusion information data packets; the data preprocessing module comprises a data filling module and a data normalization module; the data filling moduleand the data normalization module are in communication connection through a data bus; the data analysis module comprises a feature selection module and a support vector machine detection module; the feature selection module and the support vector machine detection module are in communication connection through a data bus. Missing value filling, normalization, feature selection and intrusion behavior detection are carried out on network data to achieve the purposes of high detection of the network data and more accurate judgment of the smart power grid.

The invention discloses a deep learning wafer welding spot detection method based on improved YOLOV5. The method comprises the following steps: 1) making a wafer welding spot data set; the method comprises the following steps of (1) constructing an attention mechanism module CCANET, (2) constructing a YOLOV5 network integrated with an attention mechanism, (4) introducing a Ghost module, and (5) training and improving the YOLOV5 network.According to the method, under the condition that fewer network parameters are used, the wafer welding spot detection precision is improved, and under the same condition, more wafer welding spots can be detected, and shielded wafer welding spots can also be detected.

The invention relates to the technical field of voice classification, and particularly discloses a Parkinson's disease voice data classification system based on sample and feature double transformation, which comprises a sample input module, a sample transformation module, a data set division module, a feature transformation module, a model generation module and a voting module. The system is based on the characteristic that the number of existing PD voice samples is small, and transformation is particularly carried out on two dimensions of samples and features: for sample transformation, hierarchical structures of different PD voice samples are mined through an iterative mean value clustering method, and new samples are generated; for feature transformation, PD voice feature dimension transformation is carried out through different feature kernels. The sample transformation not only can reduce the influence of abnormal samples on the classifier boundary and the influence of the samples with high correlation on the training time and the storage space, but also can reflect the hierarchical structure information of the samples in the samples. Dimension reduction is carried out on PD voice samples through feature transformation, the complexity of a classification model is reduced, and high-performance classification is achieved.

The invention relates to iris recognition, in particular to an iris recognition method based on a KNN classification model, and the method comprises the steps: determining the position of an iris, and carrying out the image preprocessing of an iris image; extracting a feature vector of the preprocessed iris image through a circular symmetry filter; carrying out dimension reduction processing on the extracted feature vectors; training a KNN classifier by using the feature vector after dimension reduction, and identifying the iris image to be identified by using the trained KNN classifier; according to the technical scheme provided by the invention, the defects that the recognition precision is difficult to guarantee and the recognition efficiency is relatively low in the prior art can be effectively overcome.