446 results about "Multi feature fusion" patented technology

The invention discloses a multi-feature fusion vehicle re-identification method based on deep learning. The method comprises five parts of model training, license plate identification, vehicle identification, similarity measurement and visualization. The method comprises the steps that a large-scale vehicle data set is used for model training, and a multi-loss function-phased joint training policyis used for training; license plate identification is carried out on each vehicle image, and a license plate identification feature vector is generated according to the license plate recognition condition; a trained model is used to extract the vehicle descriptive features and vehicle attribute features of an image to be analyzed and an image in a query library, and the vehicle descriptive features and the license plate identification vector are combined with the unique re-identification feature vector of each vehicle image; in the stage of similarity measurement, similarity measurement is carried out on the image to be analyzed and the re-identification feature vector of the image in the query library; and a search result which meets requirements is locked and visualized.

The invention discloses an Android malicious application detection method and system based on multi-feature fusion. The method comprises the following steps that: carrying out decompilation on an Android application sample to obtain a decompilation file; extracting static features from the decompilation file; operating the Android application sample in an Android simulator to extract dynamic features; carrying out feature mapping on the static features and the dynamic features by the text Hash mapping part of a locality sensitive Hash algorithm, mapping to a low-dimensional feature space to obtain a fused feature vector; and on the basis of the fused feature vector, utilizing a machine learning classification algorithm to train to obtain a classifier, and utilizing the classifier to carry out classification detection. By use of the method, the high-dimensional feature analysis problem of a malicious code rare sample family can be solved, and detection accuracy is improved.

The invention provides a text similarity measuring system based on multi-feature fusion and relates to the field of intelligent information processing. According to the system, the text similarity is measured by fusing multiple features based on word frequencies, word vectors and Wikipedia labels. The invention aims to solve the problem of semantic loss caused by non-considering of contexts in a conventional text similarity measuring system and the problem of low similarity result accuracy caused by larger text length difference. The text similarity measuring system is implemented by the following steps: carrying out preprocessing such as word segmentation and stop word removal on a training text; training corpora of the processed training text as a word vector model; measuring the similarity based on the word frequencies, the similarity based on the word vectors and the similarity based on the Wikipedia labels between input text pairs to be computed, and carrying out weighted summation to obtain a final text semantic similarity measuring result. According to the system, the measurement accuracy of the text similarities can be improved, so that the requirement on intelligent information processing is met.

A cityscape image semantic segmentation method based on multi-feature fusion and Boosting decision forest includes the following steps of carrying out super-pixel segmentation on images, carrying out multi-feature extraction, carrying out feature fusion and carrying out training learning and classification recognition. The method effectively integrates 2D features and 3D features and remarkably improves recognition rates of targets. Compared with the prior art, segmentation results are consistent, connectivity is good, edge positioning is accurate, a Boosting decision forest classification mechanism is introduced, and stability of target classification is guaranteed.

The present invention provides a wireless capsule endoscopy redundant image screening method based on multi-feature fusion. The method comprises the steps of extracting the color feature vector and the texture feature vector of an image through an HSV color feature histogram and a gray level co-occurrence matrix, a step of calculating the color feature normalized mutual information amount of an adjacent image and a texture feature mean square error value as similarity measurement, a step of considering the robustness of the method, a step of proposing the mean value method setting adaptive similarity determination threshold based on a W parameter, a step of dividing an WCE image with certain time correlation and color-texture similarity into a same sub image segment through comparing the similarity measurement value of the adjacent WCE image and a judgment threshold value orderly, and a step of using an adaptive K-means clustering algorithm to carry out key frame extraction on each sub image clip to achieve the purpose of filtering a redundant image. According to the method, the redundant image can be effectively filtered, and thus the efficiency of doctors is improved.

The invention discloses a slow denial of service (LDoS) attack detection method based on multi-feature fusion and a convolutional neural network (CNN) algorithm, and belongs to the field of network security. The method comprises the following steps: obtaining related data messages in a network key routing node in unit time to form a training sample and a test sample; Performing feature calculationon the training sample and the test sample, and generating a corresponding feature map; Using the feature map of the training sample to train a CNN model, enabling the CNN model to learn and memorizethe features of the slow denial of service attack, and finally obtaining a model which can be used for detecting the slow denial of service attack; And detecting the feature map of the test sample byusing the trained CNN model, and judging whether a slow denial of service attack occurs in a unit time corresponding to the feature map according to a judgment criterion. The detection method based on multi-feature fusion and the CNN algorithm provided by the invention can detect the slow denial of service attack in the network in a high-precision and self-adaptive manner.

The invention discloses a random ferns multi-feature fusion-based safety helmet detection method. The method comprises the steps of extracting an HSI color histogram feature, an HOG feature and an HOFfeature of a sample; then performing training for a random ferns classifier of each type and each histogram feature; according to YOLOv2 of deep learning, obtaining a human body target to serve as the to-be-detected sample; and based on fusion of random ferns classifiers of multiple features, establishing a safety helmet detection framework. According to the method, a human body target region isquickly and accurately detected based on deep learning, so that the accuracy and speed of safety helmet position locating are improved; and according to a random ferns classifier algorithm, multiple feature training classifiers are fused, so that the classification accuracy is improved. The method is simple and effective, is good in timeliness, and has very good application prospects.

The invention provides an automatic oral English marking method based on feature fusion. A multi-feature fusion method is used for marking intonation tests in a test of oral English in large scale. A continuous speech signal and a speech recognition text are used as research objects, and the intonation tests in the test of oral English are analyzed from two different angles consisting of speech and text. On the one hand, input speech is analyzed, and a speech feature is extracted and matched and compared with a reference standard; on the other hand, the speech recognition text is compared with the reading text via a text alignment method, and a marking mechanism gives assessment and a mark according to a similarity level. Experimental results show that the method is relatively low in algorithm complexity, and a marking result accords with subjective sensation of people.

The invention discloses a false comment identification method based on rolling type cooperative training, and relates to the technical field of false comment identification. The method comprises the steps: obtaining a comment text, and determining the content features of the comment text according to the comment text; obtaining reviewer information corresponding to the comment text, and determining reviewer behavior characteristics according to the reviewer information; performing false comment identification according to the comment text content characteristics and the reviewer behavior characteristics; outputting false comment recognition results. According to the false comment recognition method, unlabeled samples are effectively utilized to assist model learning, multiple features suchas emotion and text representation are fused at the same time, the recognition performance of the model is improved through multi-feature fusion collaborative training, and the accuracy is improved by 3.5% compared with a traditional false comment recognition method.

The invention relates to the field of intelligent video image processing, and discloses a face tracking method and system based on multi-feature fusion, and the method comprises the steps of obtaininga current frame target image; predicting face trajectory frame position coordinates of all stored face trajectories in the current frame of target image by using a Kalman filtering method; judging whether the frame number of the current frame of target image is a multiple of the detection interval; judging whether a human face is detected or not; performing feature extraction on the face rectangular frame by using a feature extraction algorithm to obtain face features of all faces; carrying out face feature matching; carrying out secondary matching on the face which fails to be matched and the face track which fails to be matched; updating a Kalman filter and track information; determining whether to end face tracking. By setting the face detection interval, the processing speed of the system is high; the matching accuracy is high by fusing the depth features and the HOG features; and feature matching and IOU matching are used cooperatively, so that face ID switching is reduced, the continuity of tracking is ensured, and the accuracy of tracking is improved.

The invention provides an image retrieval method based on multi-feature fusion. The image retrieval method is used for solving the problem that an image retrieval method based on a single feature cannot meet the query requirement of a user. The method comprises the steps of performing noise reduction processing on a to-be-retrieved image by utilizing a filtering method; performing feature quantification by utilizing the improved HSV color space to extract global features of the to-be-retrieved image; performing multi-scale morphological gradient extraction on the denoised image to extract local features of the to-be-retrieved image; performing adaptive fusion on the global features and the local features to obtain an adaptive fusion image; carrying out hash coding on the self-adaptive fusion image, calculating the similarity between the to-be-retrieved image and all the images in the database through Hash codes, and selecting the first several images with the highest similarity with the to-be-retrieved image as retrieval results of the to-be-retrieved image. According to the method, the feature points of the image are fully extracted, and the edge information of the image is protected more comprehensively in the local feature extraction process, so that the retrieval accuracy is improved, and the retrieval time is shortened.

The invention provides a multi-scale and multi-feature fusion feature pyramid network blind restoration method, which is used for solving the technical problems of poor restoration performance and poor robustness of an existing restoration method. The method comprises the following steps: firstly, acquiring N groups of clear images and blurred images from a data set; secondly, constructing a generative adversarial network model of a feature pyramid network and a dual-scale discriminator network; alternately training the feature pyramid network and the discriminator network by using N groups ofclear images and blurred images, and obtaining a final generative adversarial network model after a balance point is reached; and finally, inputting the blurred image to be processed into the final generative adversarial network model, and outputting a clear image. According to the method, the feature pyramid network is combined with the dual-scale discriminator, detail information loss is greatly avoided while the training speed is increased, finally, a multi-scale structure similarity loss function is introduced to further restrain image generation, and stable and high-quality restoration of the complex blurred image is achieved.

The invention discloses a mechanical rotating part performance degradation tracking method based on multi-feature fusion, and relates to a mechanical rotating part performance degradation tracking method. The objective of the invention is to solve the problems that an existing performance degradation tracking method cannot comprehensively and accurately describe all state information of a mechanical rotating part and cannot provide effective reference data for performance degradation of the mechanical rotating part. The method comprises the following steps: 1, acquiring degradation data of a mechanical rotating part; 2, extracting a plurality of features from the original vibration signal of the mechanical rotating part; 3, calculating the correlation, monotonicity, robustness and comprehensive indexes of the extracted features, and screening eight features with the highest comprehensive index value to form a sensitive feature data set; and step 4, constructing an LSTM network, inputting the sensitive feature data set screened in the step 4 into the LSTM network, and performing multi-feature fusion to obtain a fusion feature LSTM-HI which is the health factor. The method is used for mechanical rotating part performance degradation tracking.

The invention relates to a pedestrian clothing attribute recognition method based on depth attitude estimation and multi-feature fusion. The method comprises the following steps: firstly, selecting apart of retrieval results for subsequent attribute recognition through appearance feature matching; then, through an SSD-based depth human body posture estimation method, a foreground region belongingto a pedestrian in the image can be effectively positioned, and background factor interference can be well eliminated; and finally, fusing analysis results of multiple modes, and combining an iterative smoothing process to enhance the correlation between an attribute tag and a pixel by adopting a maximum posterior probability distribution mode to obtain a final attribute analysis recognition result. According to the method, the problems of inaccurate label recognition, pixel analysis area deviation and the like under a single analysis mode are solved. The method is simple and flexible, and has high practical applicability.

The invention provides an image segmentation method for enhancing edge and detail information by utilizing feature fusion, and relates to the technical field of computer vision. The method comprises the steps of extracting features of an input image by using a convolutional neural network; inputting the extracted features into a decoding structure added with more feature fusion, and enriching theedge and detail information while recovering the image resolution to obtain a dense feature map; outputting maximum values of different classifications through a normalization method; and calculatinga cross entropy loss function, and updating the weight in the network by using a random gradient descent method. According to the method, the position and boundary detail information lost in the encoding stage can be recovered while the resolution of the feature map is recovered, the information of the picture is enriched, the dense feature map is obtained, the sparse feature map brought by directup-sampling is compensated, the segmented boundary and details are clearer, and the segmentation effect on detail small objects is improved.

The invention relates to a shaking detection algorithm based on multi-feature fusion. The algorithm comprises the following steps of: initializing an undistorted video frame and a video frame to be detected; then, sequentially extracting a brightness similarity factor, a contrast similarity factor and a structure similarity factor to obtain a similarity metric function, wherein the contrast similarity factor is obtained after the average brightness is removed; and meanwhile, judging that the video frame of a signal to be detected does not shake, or judging that the video frame of the signal to be detected shakes. According to the shaking detection algorithm, the brightness similarity factor, the contrast similarity factor and the structure similarity factor of the video frame of a reference signal and the video frame of the signal to be detected are sequentially extracted, the similarities of the three kinds of extracted variable information are compared, and finally, the three comparison results are combined, so that a similarity index is obtained and used as an evaluation criterion for image quality, i.e., indicating whether shaking happens or not.

The invention discloses a particle filter target tracking method based on multi-feature fusion. The method comprises the following steps: acquiring a video image and performing filtering processing; calibrating a tracking target in the initial frame by using a rectangular frame, and calculating an edge histogram, a texture histogram and a depth histogram of the target template; updating a particlestate by adopting a second-order autoregression model, and obtaining a feature histogram of each particle; calculating the similarity of the two templates, obtaining a single feature discrimination degree according to the position mean value, the standard deviation and the overall position mean value of the particles under the single feature, and adaptively adjusting the fusion weight; determining the particle weight at the current moment in combination with the observation model of multi-feature fusion and the particle weight at the previous moment; sorting the particle weights, counting thenumber of particles with small weights, comparing the number of particles with small weights with a threshold value, correcting the size of a window, and determining the state of a tracking target. According to the method, the edge, texture and depth characteristics are combined, so that the target can be tracked more accurately and continuously.

The invention relates to a tunnel fire early warning method based on image processing, and the method comprises the steps: obtaining tunnel video stream data, and carrying out the preprocessing of thedata, and the preprocessing comprises graying and smoothing; segmenting the preprocessed data to obtain a segmentation result, the segmentation result comprising a flame motion foreground; flame motion features are extracted from the flame motion foreground, the flame motion features comprise flame shape features, flame texture features and flame area features, and flame color features are extracted from the tunnel video stream data; and performing multi-feature fusion on the flame color feature, the flame shape feature, the flame texture feature and the flame area feature, performing flame prediction through a pre-trained neural network, and outputting fire early warning information based on a flame prediction result. According to the invention, the fire identification accuracy can be improved.

A multi-feature fusion-based polarimetric SAR image classification method disclosed by the present invention comprises the steps of firstly extracting the polarimetric feature vectors of a to-be-classified polarimetric SAR image to obtain a high-dimension polarimetric feature set; extracting the morphological profile feature vectors of the SPAN processing results of the image to obtain a high-dimension morphological feature set; carrying out the dimension reduction processing of the locality preserving discrimination analysis on the two kinds of high-dimension features separately, and then selecting the image pixel points of the known category labels to form a training sample set, and then selecting the rest pixel points of the whole image as a classification sample set; using a maximum posterior probability-based support vector machine (SVM) to process the two kinds of low-dimension features separately to obtain the category labels and the corresponding posterior probabilities of the pixel points on the respective conditions; adopting a summation criterion or an adaptive weighted summation criterion to combine the posterior probability vectors of each pixel point on the two conditions, and according to the maximum posterior probability principle, obtaining a final classification result of the high-resolution polarimetric SAR image. According to the present invention, the classification accuracy and the efficiency of the resolution polarimetric SAR image are helpful to be improved.

The present invention relates to an SVM mammary gland classification method based on Gaussian kernel parameter selection. The method comprises: extracting a mammary gland molybdenum target and B ultrasonic image features from a known case, and performing benign and malignant and clinical stage mark of each case data which completes the feature extraction according to the known clinical diagnosis result; performing multi-feature fusion of the molybdenum target image features and the B ultrasonic image features of the same patient's mammary gland through adoption of the cascade mode, and obtaining the feature vector of mammary gland samples; and allowing the Gaussian kernel parameter selection method to be used for the training process and the identification process of the dichotomy balance decision tree SVM multi-classification algorithm based on the Gaussian kernel. The accuracy and the efficiency of the breast cancer diagnosis can be improved.

The invention discloses a BERT-based multi-feature fusion fuzzy text classification model. The method comprises the following steps of preparing a fuzzy text classification original data set; a BERTMFFM model is constructed, the BERTMFFM model comprises a BERT model, a convolutional neural network, a bidirectional long-short memory network and a SelfAttention module, the input of the BERT model is a fuzzy text, the output of the BERT model is connected with the convolutional neural network, the bidirectional long-short memory network and the SelfAttention module, and local features, sentence semantic features and syntax structure features of the fuzzy text are extracted; splicing the output of the BERT model with the output of the bidirectional long-short memory network at the same time, and then screening out optimal sentence semantic features by using maximum pooling operation; and fusing the local features, the optimal sentence semantic features and the syntactic structure features by adopting a parallel splicing mode, and performing fuzzy text classification on a fusion result through a SoftMax function to finish the construction of the BERTMFFM model. The problem of incomplete feature collection is solved, so that the classification accuracy is improved.