42results about How to "Reduce the probability of misclassification" patented technology

The invention discloses an auroral oval segmenting method based on a brightness self-adaptive level set, which mainly solves the defects of the existing auroral oval segmenting method that the segmentation precision is low, the robustness is poor and the application range is small. The auroral oval segmenting method comprises the following steps of (1) adopting a morphology component analysis method to preprocess an ultraviolet aurora image; (2) establishing a morphology saliency map to be used as shape characteristics of the auroral oval; (3) utilizing the marginal curve of the morphology saliency map to initialize a level set function; (4) calculating the brightness self-adaptive level set evolution speed and a stop function; (5) updating the level set function according to the brightness self-adaptive level set evolution equation; and (6) extracting a zero level set curve after being updated and utilizing the zero level set curve as the auroral oval margin to be outputted. Due to adopting the auroral oval segmenting method, the phenomenon of the traditional segmenting method such as result deviation and margin leakage can be avoided, advantages such as high segmentation precision and strong robustness can be achieved, and the method is applicable to the segmentation of different ultraviolet auroral images.

The invention relates to an unsupervised Markov random field image segmentation method, which is technically characterized in that: the scope of the classification number of an image M is firstly determined, the image M is then carried out the K-means segmentation from m min to m max type, the BICm which is corresponding to the image M under each type is calculated according to the BIC criteria, a potential energy function is applied to calculate the total energy of the image M, and finally the ICM method is selected to complete the Markov random field image segmentation. The method has the advantages that: the new potential energy function is applied to image segmentation, the anti-noise performance thereof is obviously improved, the segmentation result can remove most of the noise of the image with great noise, the error classification phenomenon can be simultaneously effectively reduced, the error classification rate is reduced by more than 60 percent; the classification number does not need to be manually determined, thus basically realizing the unsupervised image segmentation; and the judgment accuracy of the classification number can achieve more than 95 percent by the artificial synthetic image test.

The invention belongs to the technical field of machine vision, and concretely relates to a metal part surface defect detection and identification method as well as an apparatus thereof, which can solve the problem of low detection and identification precision under different environments in the prior art. The metal surface defect detection and identification method is characterized in that basedon an original image of the surface of a metal part, a surface defect detection model is constructed in advance for performing defect detection on the original image to obtain a defect area position;the defect area position is subjected to contour detection to respectively obtain a first defect contour and a second defect contour; the first defect contour and the second defect contour are input in the pre-constructed surface defect detection model, and the defect types of the surface of the metal part can be identified. The method has the advantages of full automatic performance and high detection precision.

The invention discloses an improved support vector machine-LIBS (laser-induced breakdown spectroscopy) combined sorting method for steel materials. The method comprises the steps of detecting a series of rolled steel samples with known marks with an LIBS system to obtain data matrixes of the rolled steel with different marks, and utilizing a support vector machine to establish a sorting model on the known category data, wherein in the modeling process, an improved modeling method, namely combined modeling is used; after data of the sample to be measured is input to the model, performing fuzzy classification with a one-to-many method, screening candidate categories, and performing sophisticated classification by a one-to-one method, and finally determining the category of the data to be measured. According to the method, the one-to-many and one-to-one modeling methods are combined for use, and advantages of the methods are fully utilized, so the data to be measured can be subjected to two layers of analyzing systems of the fuzzy classification and the sophisticated classification, the influence of useless category information on a prediction process is reduced, the prediction accurate rate can be obviously improved and the calculating cost is reduced.

The invention relates to the technical field of ship detecting and tracking, in particular to a ship detecting and tracking method based on an infrared video. The method comprises the following steps: detecting and extracting a sea-sky line through a gradient operator method and a Hough transformation method, and segmenting a sea area from a video picture using the sea-sky line; extracting the foreground of an infrared video based on an improved ViBe algorithm, extracting a suspected target from the sea area of the video, and repairing an extracted image; analyzing the texture features of the image based on a statistical histogram, and capturing the target; and tracking the moving target through a MeanShift algorithm based on multiple features and saliency detection. The video picture is segmented, search is carried out in the sea area alone, and the search scope is reduced. Through background subtraction based on the improved ViBe algorithm, the influence of island and shore background can be eliminated. By capturing the target based on a statistical histogram, the accuracy of target capture is improved. Target tracking based on the MeanShift algorithm is fast.

The invention relates to a reeling steel state identifying method based on KNN reeling steel picture data of an optimized KAP sample, wherein KAP belongs to a fast efficient clustering algorithm, clustering is performed by the mutual relation among computation features, a new sample space is formed by clustering and selecting feature samples with representation in various samples, and the balance among each sample number in the new sample space is ensured by fixing the parameter K in the KAP algorithm. A PCA algorithm is utilized for performing dimension reducing on the features and the operation speed of the method is improved under the condition of ensuring the classifying precision. The classifying identification on the linear array CCD (charge couple device) picture-based reeling steel loading state under the complex natural environment is realized. The method has very high classifying precision and lower wrong classifying rate, and the actual needs are met with the faster classifying speed.

The invention belongs to the field of video processing and pattern recognition, and specifically relates to a video-based railway wagon dropped object automatic detection method through moving target detection based on background modeling and a target recognition algorithm based on a neural network. The moving target detection based on background modeling and the target recognition algorithm based on a neural network comprise the following steps: S1, acquiring a video; S2, building a train running gear histogram gradient feature model; S3, extracting a suspected dropped object by use of a background modeling technology; and S4, identifying the suspected dropped object based on an artificial neural network (ANN). Through the method, an object dropping off from a wagon can be detected and identified automatically under high-speed running of a train, high detection accuracy and low error rate are achieved, and the high detection speed meets the needs of practical engineering.

The invention discloses a spectrum sensing method in a cognitive wireless network environment. The method comprises the following steps: a data set is constructed, and a secondary user in the cognitive wireless network detects the primary user; each secondary user collects the energy value in the channel and sends the energy value to the fusion center, the fusion center takes the energy values sent by all the secondary users as feature vectors, the primary user in the cognitive wireless network discontinuously sends the working state of the primary user to the fusion center, and the fusion center takes the working state sent by the primary user as a data set label; the data set is trained through an Xgboost algorithm to obtain a machine learning model of a current channel sensing result; after the trained model is obtained, the secondary user continues to collect the energy value in the channel and sends the energy value to the fusion center as test data, so that the Xgboost algorithmmodel of the fusion center performs spectrum sensing in the cognitive radio network environment. According to the method disclosed in the invention, the performance deficiency of a machine learning algorithm in the spectrum sensing problem in the prior art is improved.

The invention discloses an anti-attack defense method based on a neuron activation rate. The method comprises the following steps: 1), obtaining an image data set, carrying out the preprocessing, and constructing a training set and a test set; 2) constructing a classifier model of a CNN network structure; 3) constructing a dichotomy discriminator D, and judging the output N of the classifier model about the input book x; 4) constructing an overall loss function according to the classifier model and the loss function of the dichotomy discriminator D, and determining an overall optimization target; carrying out joint training on the overall model by adopting the training set, and when the overall loss function is converged, stopping training and storing the model; and 5) inputting a test set to test the trained classifier model, if the classification accuracy is lower than a threshold value, modifying the lambda of the overall loss function, and training again according to the step 4) until the classification accuracy is higher than the threshold value. The defense method provided by the invention has good applicability and accuracy, and has a good defense effect on attack resistance.

The invention provides a spectral angle mapping method used for correcting the negative correlation by wavebands, which is technically characterized in that: the traditional spectral angle is calculated, the value of the newly added spectral bands is used as the independent variable, the existence of the negative correlation on each newly added spectral band is judged, and the newly added spectral band with the negative correlation is provided with correction parameters. The spectral angle mapping method provided by the invention aims to solve the problem that the traditional spectral angle mapping method can not distinguish the negative correlation of spectrums so that the spectral curves with different characteristics are classified into the same category relative to a certain reference spectrum. The experiments proves that: due to the adoption of the spectral angle mapping method provided by the invention, the separability of the spectrums is improved effectively, and the spectral vectors which can not be separated by using the traditional spectral angle mapping method can be separated by stages according to the difference of the generated negative-correlation wavebands. The spectral angle mapping method provided by the invention laids a foundation for the subsequent hyperspectral remote sensing image classification, the target identification and the like.

The invention discloses a depression state detection system based on lateral brain habenula nucleus signals. The system comprises a computer memory, a computer processor and a computer program which is stored in the computer memory and can be executed on the computer processor, and a trained depression state detection model is stored in the computer memory. The depression state detection model is constructed by using a plurality of groups of different types of weak classifiers, and each weak classifier selects part of feature values from sub-feature spaces constructed in a training set and a test set for training. When the computer processor executes the computer program, the following steps are realized: acquiring a lateral brain habenular nucleus signal, inputting the lateral brain habenular nucleus signal into a trained detection model, extracting a high-dimensional feature vector of each time slice, counting depression and non-depression labels deduced by each weak classifier, and generating a prediction label corresponding to each time slice by adopting a majority voting system. The method can improve the detection precision, and is of great significance to clinical diagnosis, evaluation and treatment of depression.

The present invention discloses a polarization SAR image classification method based on non-stationary modeling of a polarization scattering mechanism, in order to solve the problems that the existingpolarization SAR image classification is affected by noise and has low accuracy for the mixed pixels with no obvious main scattering mechanism. The implementation steps are: initially classifying measured images; estimating the auxiliary random field according to the polarization scattering characteristics, and associating the polarization scattering characteristics with the non-stationarity; dividing the pixel point stationarity by using the auxiliary random field; calculating correlation functions for the stationary pixel points to obtain a unitary potential energy function, a data item, and a binary potential energy function; calculating the membership degree for non-stationary pixel points; constructing a posterior probability model of a fuzzy triple-recognition random field (FTDF) model by using the obtained functions, and performing classification by using the maximum posterior probability criterion; and if it is marked that the random field converges, outputting a result, and otherwise repeatedly constructing the classification model according the iterative rule until the termination iteration requirement is reached, and outputting a classification result. The method disclosed by the present invention has high detection precision and good anti-noise performance, and can be used for polarization SAR image classification.

The invention relates to a multitask text analysis method based on text sentiment analysis of CNN-LSTM and textrank abstract automatic extraction of word2vector. The method comprises the steps of obtaining massive to-be-tested network text data, firstly, preprocessing network text data to be tested and then inputting the preprocessed network text data into an LSTM-CNN neural network; according tothe LSTM-CNN, a classical text sequence processing method being used for a long-term and short-term memory network; obtaining a vector representing the context; the CNN further extracting higher-dimensional and effective features; then, sending features into softmax to be subjected to multi-classification, so that sentiment positive and negative directions of a text are obtained, secondly, segmenting the input text data into sentences by combining a textrank algorithm based on word embedding to construct a graph model, and calculating the similarity between the sentences to serve as weights ofedges; by calculating sentence scores, sorting the obtained sentence scores in an inverted order, and extracting several sentences with the highest importance degree as candidate abstract sentences;finally, displaying the analysis result in the form of a report. The multi-task text data processing model enables a public opinion monitoring result to obtain high accuracy and high efficiency, and text analysis precision is improved by using two neural network training.

The invention discloses a dynamic classifier chain adjusting method for multi-label classification, belonging to the field of machine learning field. The technical problem to be solved by the invention is how to reduce the probability that the output is wrongly divided in eyes near a threshold value, the randomness of a mark prediction sequence existing in a classifier chain is relieved; uncertainty and instability of classification results are realized; the adopted technical scheme is as follows: the device comprises a base, the method is characterized in that training data are concentrated;respectively counting the co-occurrence frequency of each mark and the marks except the mark, and progressively decreasing and sorting; the method comprises the following steps: selecting a classifierfrom a training data set to complete the sorting of classifier chains in the training data set, randomly selecting one classifier from the classifier chains to complete the classification of unknownsamples, setting two thresholds in advance during the classification of the unknown samples, and completing the classification of the unknown samples according to the output values of the randomly selected classifiers and the sizes of the two thresholds.

The invention discloses a stable evaluation method for electronic blood glucose measurement, which belongs to the field of medical measurement devices. The method comprises the steps of monitoring the variation of current in the reaction process in real time by utilizing an electrochemical analyzer, extracting n sample current values under the situation that the detection current is stable, getting an average value, obtaining a real-time amplified voltage value through a voltage amplifier, and obtaining a measurement value of the blood glucose through an analog-digital conversion and functional conversion method. By adopting the method, the error of the blood glucose electronic measurement caused by uncontrollable factors can be reduced, so that a stable and precise blood glucose value can be obtained.

The invention discloses a cooperative energy detection spectrum sensing method based on a Lightgbm algorithm. The method specifically comprises: in a cognitive wireless network, a secondary user detecting energy in a current channel environment and sending a result to a secondary user serving as a fusion center, and a primary user discontinuously sending a label indicating whether spectrum resources are used by the primary user or not to the fusion center; establishing a data set by using an energy detection-based method, wherein the Lightgbm algorithm specifically uses a gradient-based unilateral sampling technology and a unique feature binding technology; dividing the constructed energy feature vector set into a training set and a test set, and performing training and testing respectively; the fusion center distinguishing whether a channel is available or not, and informing all secondary users after a result is obtained. According to the method, under the condition that the false alarm rate 0.1 meeting the IEEE802.11 requirement is used, the detection rate is increased by 6%-7% compared with an SVM, and meanwhile the false classification rate and the false classification risk areremarkably reduced.

The invention relates to an image-based object classification method and system and an electronic device. The method comprises the steps of identifying a first area where an object of an image is located; generating a plurality of second regions on the object characteristics of the first region; subdividing the plurality of second regions to obtain a plurality of second subdivided regions; classifying and screening out a third area containing the to-be-detected object, subdividing the third area to obtain a fourth area containing the distinguishing characteristics of the to-be-detected object,and finally identifying and classifying the first area, the second area, the third area and the fourth area where the to-be-detected object is located to obtain the classification information of theto-be-detected object. The accuracy of object classification is improved by carrying out multi-stage classification and screening on the object areas of the to-be-processed image.