64 results about "Support vector machine svm classifier" patented technology

Disclosed is a multi-camera system target matching method based on a deep-convolution neural network. The multi-camera system target matching method based on the deep-convolution neural network comprises initializing multiple convolution kernels on the basis of a local protective projection method, performing downsampling on images through a maximum value pooling method, and through layer-by-layer feature transformation, extracting histogram features higher in robustness and representativeness; performing classification and identification through a multi-category support vector machine (SVM) classifier; when a target enters one camera field of view from another camera field of view, performing feature extraction on the target and marking a corresponding target tag. The multi-camera system target matching method based on the deep-convolution neural network achieves accurate identification of the target in a multi-camera cooperative monitoring area and can be used for target handoff, tracking and the like.

The invention discloses a pedestrian detection method and a pedestrian detection device. The method comprises the following steps of: collecting a pedestrian image, preprocessing the image, and extracting a region of interest from the preprocessed image; pre-positioning a pedestrian region in the region of interest by adopting an Adaboost cascade classifier; and positioning the pedestrian region in the pedestrian region pre-positioned by the Adaboost cascade classifier by adopting a support vector machine (SVM). By the technical scheme, the technical problem that the fallout ratio in the prior art is high.

The invention discloses a video-analysis-based smoke detecting method, which is realized by taking a digital camera as a sensor and under support of a digital signal processing chip. The method is characterized by comprising the following steps: collecting digital video by utilizing the digital camera; screening out a foreground part containing a movement part; screening out an area with likeness by using a support-vector-machine detector; analyzing high-frequency signal change by applying wavelet transform, and screening out a digital image of which the background is in gradually fuzzy change; and screening out digital images with smoke texture characteristics by using an Adaboost cascade classifier pairs. By combining moving object extraction by background modeling, grey area screening by a support-vector-machine (SVM) classifier, high-frequency signal change analysis by wavelet transform, the smoke texture characteristic cascade classifier and other methods, the method greatly improves the accuracy and efficiency of smoke detection, reduces the false alarm rate and has high robustness and higher practicality.

The present invention relates to a diabetic retinopathy sign detection method and device. The method comprises: receiving a fundus image to be detected; using a convolutional neural network (CNN) model to perform processing of the fundus image to be detected, and obtaining lesion area samples corresponding to the fundus image to be detected; according to the lesion area samples, constructing SIFT (Scale-Invariant Feature Transform) feature descriptors corresponding to the fundus image to be detected; and according to the SIFT feature descriptors, using an SVM (Support Vector Machine) classifier to determine the lesion type of the fundus image to be detected. The CNN model is employed to perform rough classification of the diabetic retinopathy to five out a lesion area, the SVM classifier is employed to perform fine classification of the lesion type of the lesion area to reduce interference generated by difference between the data size and the data when the neural network is directly configured to perform lesion classification and improve the classification precision.

The invention provides a software defect prediction method and system. The software defect prediction method and system are used for resolving the problem that existing software defect prediction is low in accuracy. The software defect prediction system comprises a dimension reduction processing unit, an SVM training unit and a defect prediction unit. The software defect prediction method comprises the steps that firstly, dimension reduction processing is conducted on a first training dataset according to the LLE, a lower-dimension vector, mapped into a low-dimension space, of each sample point in the first training dataset is obtained, and a second training dataset composed of the lower-dimension vectors is obtained; secondly, training is conducted on an SVM classifier according to the second training dataset, an optimal classification hyperplane function of the SVM classifier is obtained, and the trained SVM classifier is further obtained; thirdly, detect prediction is conducted on software to be predicted according to the trained SVM classifier.

The invention discloses a face detecting method and device. Face detection is performed based on a method combining face detection and face alignment. The face detecting method includes scanning to-be-recognized images line by line and row and row and judging each input window is a face window or not; combining all the face windows and obtaining a final face area on the original to-be-recognized images, wherein during a process of judging whether each input window is a face window or not, an Adaboost frame based cascade classifier is used, pixel difference is adopted as extraction characteristics, a random forest classifier acts as a weak classifier for face detection and an irritation algorithm for face alignment and an SVM (Support Vector Machine) classifier is adopted for secondary judgment, and the windows pass the judgment are the face windows. By adopting the above scheme, the invention provides the face detection method and device with accuracy and speed capable of meeting requirements.

The invention discloses an efficient Android malware detection model DroidDet based on rotating forest and belongs to the technical field of information security. According to the invention, a decompilation tool is used to decompile the Apk file in an Android application program and a combination of four groups of features is analyzed and extracted as the input for building a rotating forest classifier model. The model is characterized in that 1) the integrated learning method-rotating forest algorithm is initially adopted; further, each base learning device is trained through adopting PCA major constituent analysis and self-service sampling method and using a whole training set; accordingly, the model created in the invention has superior generalization performance; 2) the rotating forest algorithm has the feature of selecting the optimal characteristics so as to avoid noise data. The 10-fold cross-validation method is used for verifying the validity of the model created in the invention and the predicting accuracy of the model in the invention reaches 88.26%, which is 3.13% higher than the 84.93% of typical support vector machines. The model of the invention has wide application prospect in user information security.

The invention belongs to the field of pattern recognition, and particularly relates to a car logo locating and recognizing method and a system. According to the car logo locating and recognizing method and the system, car logos can be recognized. The car logo locating and recognizing method includes the steps: collecting pictures of samples of various car logos, and obtaining a template base of the car logos; adopting a fast algorithm, and extracting out a feature set of the sample picture of each car logo in the template base of the car logos; serving the feature sets as a training set, and generating a support vector machine (SVM) classifier. The method further includes the steps: collecting original images of the car logos; carrying out morphology preprocessing on the original images, and locating a candidate area of the car logos on images obtained through the morphology preprocessing; adopting the fast algorithm, and extracting out a to-be-recognized feature set of the candidate area of the car logos; adopting the SVM classifier, carrying out matching recognition on the to-be-recognized feature set, and serving a car logo which corresponds to a feature set matched with the to-be-recognized feature set as a recognition result.

The invention belongs to the field of power technology and computer vision, and aims at separating a normal image from an image with a nest, locating and finding a problem more quickly and meeting thedemands of the construction of an intelligent power grid. The invention discloses a method for recognizing a nest on a power transmission line based on HOG features and machine learning. The method comprises the following steps: 1, HOG feature extraction of a directional gradient histogram; 2, principal component analysis; 3, training of an SVM (support vector machine) classifier: 1) performing normalization; 2) extracting a feature vector of a training set obtained at a former step to form a training set of the classifier, and making a label file conforming to an SVM format; 3) finding the optimal parameters through testing; 4, the inputting of a test set image, classification through the trained classifier, and outputting of a final classification result. The method is mainly used in occasions of automatically recognizing the nest faults of the power equipment through images.

The invention discloses a hyperspectral image classification method based on K nearest neighbor filtering. The classification process mainly includes (1) support vector machine (SVM) classification: rough classification of a hyperspectral image using a SVM classifier to obtain an initial probability graph; (2) principal component analysis dimensionality reduction: dimensionality reduction of the hyperspectral image by way of principal component analysis to obtain a first principal component image; (3) K nearest neighbor filtering: extraction of spatial information of the hyperspectral image under the guidance of the first principal component image based on a non local K nearest neighbor filter to optimize the initial probability graph; and (4) accurate classification of the hyperspectral image according to the optimized probability graph. The greatest advantage of the method in the invention over a traditional hyperspectral classification algorithm is that the non local spatial information of the hyperspectral image can be extracted for optimized classification without solving for a complex global energy optimization problem. Thus, the classification speed is high, and the accuracy is high.

The invention provides a flame image recognition method and device and a storage medium, and relates to the technical field of image recognition. The flame image recognition method comprises the following steps: determining a suspected flame region in an acquired image; extracting a texture feature image of the suspected flame region by adopting a filter; extracting a local binary feature of the texture feature image based on a local binary pattern; and determining whether flame exists in the suspected flame area or not based on the local binary characteristics by adopting a support vector machine (SVM) classifier. According to the method, through extraction of the texture feature image and further local binary features, the flame recognition efficiency and accuracy are improved.

The invention relates to a breast tumor classification algorithm based on a convolutional neural network VGG16. The algorithm comprises the following steps that: data preprocessing: for a dataset which presents a data imbalance state, carrying out imbalance processing and data enhancement processing; the establishment of the convolutional neural network: 1) network pre-training: utilizing the VGG16 to carry out network training on an ImageNet large natural image dataset, and storing trained weight; 2) network key node selection: utilizing different layers of the VGG16 network to carry out feature extraction on a breast tumor DDSM (Digital Database for Screening Mammography) dataset, applying the same SVM (Support Vector Machine) classifier for classification for extracted features, and selecting a layer with highest classification performance as a node constructed by a new network; and 3) connecting two layers of full connection and one layer of softmax to form a new network behind thenode constructed by the selected network; and carrying out migration learning.

The invention discloses an aurora image sequence classification method based on a space-time polarity local binary pattern. The aurora image sequence classification method based on the space-time polarity local binary pattern mainly resolves the problem in the prior art that the classification efficiency is not high. The aurora image sequence classification method comprises the steps that (1) preprocessing of rotating an aurora image sequence counterclockwise by 62.63 degrees is carried out; (2) the preprocessed aurora image sequence is chunked multiple times, and the numbers of chunks of the multiple times are different; (3) a polarity local binary pattern PVLBP algorithm is used, the polarity local binary pattern characteristic PVLBP of each chunk is extracted, and the PVLBP characteristics of all the chunks are sequentially connected to obtain a space-time polarity local binary pattern characteristic ST-PVLBP; (4) the space-time polarity local binary pattern characteristic ST-PVLBP of the aurora sequence is input into a support vector machine (SVM) classifier to obtain a classification result. The aurora image sequence classification method based on the space-time polarity local binary pattern keeps high classification accuracy, shortens the classification time, improves classification efficiency, and can be applied to scene classification and event detection.

The invention discloses a message anomaly detection method based on deep learning. According to the method, firstly, data is preprocessed into a two-dimensional matrix form, and in order to reduce anoverfitting phenomenon easily occurring in a general algorithm model, a shuffle function is used for randomly disturbing the data to prevent local optimum, so that the model is easier to converge; then effective features are learned from the preprocessed data by using a convolutional neural network (CNN), and finally classification processing is performed on the obtained vectors through a supportvector machine (SVM) classifier. According to the method, the advantage of high-dimensional learning of the convolutional neural network of deep learning is utilized to learn the effective features ofthe network transmission message; the model provided by the invention is subjected to experimental verification on a Beijing University data set and is subjected to experimental comparison with othertwo models with better effects verified on the data set, experiments prove that the accuracy and the stability are greatly improved, and the training and testing time consumption is obviously reduced.

The invention relates to a fault diagnosis method and system of a planetary gear transmission system, belonging to the fields of a fault diagnosis technique and a signal processing and analyzing technique. The fault diagnosis method comprises the following main fault diagnosis steps: (1) processing acquired vibration signals of the planetary gear transmission system by utilizing a homomorphic filtering technical method, and separating signals comprising fault features and external excitation background noise signals to reduce the complexity of data; (2) carrying out frequency spectrum reconstruction on the signals comprising the fault features after homomorphic filtering processing by adopting complex cepstrum transform, and extracting the fault feature information in the vibration signals of the planetary gear transmission system; and (3) diagnosing the running state of the planetary gear transmission system to be the normal state or the fault state by combining a vibration mechanism of the planetary gear transmission system, utilizing a classifier of a support vector machine (SVM) and taking the time domain statistics at different states and peak value energy after frequency spectrum reconstruction as fault feature vectors.

The invention provides a character recognition method and device. The character recognition method is applied to recognition for a character in an image. The image comprises a text box with the position being uncertain, and the character is filled in the text box. The character recognition method comprises the steps of determining the position of the text box in an image to be recognized; intercepting a target image corresponding to the text box in the image to be recognized according to the position of the text box; calculating a target histogram of oriented gradient (HOG) feature description operator corresponding to the target image; acquiring a character to be recognized of the target image according to the target HOG feature description operator and a pre-trained support vector machine (SVM) classifier. By applying the embodiment of the invention, the position of the text box is determined in the image to be recognized, and a target image corresponding to the text box in the image to be recognized is intercepted according to the position of the text box, so that active positioning and interception for the target image are realized, an error occurred when a target image area is intercepted according to preset fixed coordinates is avoided, and the recognition accuracy is improved.

The invention provides a 3D (Three-dimensional) gesture recognition method and system. The 3D gesture recognition system comprises a feature extracting unit, a matching unit and a support vector machine (SVM) classifier, wherein the feature extracting unit is used for extracting a grid depth feature (GDF) from a present frame of an input video sequence, and extracting level setup moment (LSM) features and/or curvature histogram (HOC) features; the matching unit is used for matching the GDF feature extracted by the feature extracting unit with GDF features of a plurality of clustering templates obtained from off-line view clustering, thus obtaining orientation information of a gesture in the present frame; and the SVM classifier is used for recognizing the gesture in the present frame based on the LSM features extracted by the feature extracting unit and/or the HOC features and the orientation information obtained by the matching unit. According to the 3D gesture recognition method and system, the orientation of a hand is not limited, and the problem that the hand of an actor may be shielded by the actor is solved successfully.

The application provides an abnormal motion detection method and device. The method comprises the steps that foreground objects in a monitoring video are detected according to depth information; depth difference of the foreground objects between adjacent frames is calculated so that depth difference images are obtained; continuous multiple frames of depth difference images are calculated so that polymerized depth difference images are obtained; features of a histogram of oriented gradients HOG are calculated according to the polymerized depth difference images; and abnormal motions corresponding to the HOG features are predicted by a support vector machine SVM classifier with well grained abnormal motions, and whether the abnormal motions occur on the foreground objects is confirmed according to the prediction result. According to the scheme of the embodiment of the application, the foreground objects in the monitoring video are detected according to the depth information so that people with different distances to a lens of the same position in the frame can be accurately separated, and thus the abnormal motions of each person in the scene can be accurately judged.

The invention discloses a target detection method. The method comprises the steps of acquiring a feature graph and a double resolution feature graph of an input image; a root filter of a deformable part model acting on the feature graph and capturing a global feature of a target object, and all part filters of the deformable part model acting on the double resolution feature graph and capturing a local feature of the target object; a plurality of support vector machine (SVM) classifiers based on expectation maximization (EM) algorithm training identify the target object in the input image according to the global feature and local feature of the target object; and a detection result is output. Correspondingly, the invention also discloses a target detection system. The detection precision and performance of the system can meet actual requirements, and furthermore, large scale industrial application of the system is expected to realize in a short term.

The invention discloses a palm detection method for a palm image of a computer system. The method comprises the steps of graying a palm image by a trigger command system; extracting the features of a gradient orientation histogram (HOG) of the above grayed palm image; and classifying the features of the HOG of a target palm image by means of a trained support vector machine (SVM) classifier. The trained SVM classifier is obtained in advance based on the classified training process of images containing valid palm patterns and images containing invalid palm patterns. Based on the above method, palm patterns in palm images can be effectively detected.

The invention belongs to the computer vision technical field and provides a target feature detection method and device. The method comprises the following steps that: a LBP (Local Binary Pattern) feature descriptors are extracted from a video frame; according to a preset simplification algorithm, a HOG (Histogram of Oriented Gradient) feature descriptors are extracted from the video frame; and a target can be identified through an SVM (Support Vector Machine) classifier according to the LBP feature descriptors and the HOG feature descriptors. According to the target feature detection method and device provided by the invention, the LBP feature descriptors and the HOG feature descriptors of the video frame are extracted according to the preset simplification algorithm so as to form HOG_LBP feature descriptors, and the HOG_LBP feature descriptors are inputted into the SVM classifier, so that a specific target in the video frame can be identified. The target feature detection method is suitable for FPGA (Field Programmable Gate Array) hardware circuits, and therefore, requirements for real-time detection can be satisfied, and computing speed of target feature detection can be improved.

The invention discloses an efficient spectrum sensing method based on a support vector machine. The efficient spectrum sensing method comprises the following steps: S1, inputting a to-be-sensed receiving signal; S2, preprocessing the to-be-perceived received signal through PCA (principal component analysis), and decomposing a covariance matrix of the to-be-perceived received signal by adopting Dullet decomposition to obtain feature statistics; S3, obtaining a label of a to-be-perceived received signal through an energy detection algorithm, and forming a sample training set by the obtained label and the obtained feature statistics; S4, inputting the formed sample training set into a support vector machine SVM classifier for training to obtain a spectrum classifier; and S5, inputting the collected data into a spectrum classifier for processing to obtain a classification result. According to the method, the high spectrum recognition rate can still be achieved under the condition of the low signal-to-noise ratio, meanwhile, due to introduction of the non-progressive threshold, the progressive threshold changes along with the environment, and spectrum sensing is more accurate.

The invention discloses a feature fusion-based vehicle behavior recognition method in an urban traffic scene. An HOG feature and an LBP feature are in head-tail serial connection to form a joint feature as a vehicle behavior fusion feature, a SVM (support vector machine) classifier is adopted, and vehicle behaviors are classified and recognized. According to the feature fusion-based vehicle behavior recognition method in the urban traffic scene disclosed by the invention, abnormal vehicle behaviors can be recognized, a frontier foundation is laid for an automatic traffic event detection system, and an important role is played in improving the vehicle driving safety and avoiding a road traffic accident.

The invention provides an emotion recognition method and an emotion recognition device for customer consultation texts and belongs to the technical field of date service. The emotion recognition method for customer consultation texts comprises choosing characteristics of a customer consultation text training set to form a characteristic set which comprises a flagged text having being flagged emotion category; converting the flagged text into a characteristic vector represented in the characteristic set to obtain a training data set; training the training data set to generate a support vector machine (SVM) classifier; and inputting the to-be-analyzed customer consultation texts into the SVM classifier to be analyzed to obtain emotion categories represented by the customer consultation texts. The emotion recognition method can improve emotion classification accuracy.