46results about How to "Reduce classification error" patented technology

The invention discloses a time-space condition information based moving object detection method. The method comprises the following steps: building a target detection time-space domain model through considering the significance of human visual time-space domains; calculating a conditional probability that a detection image belongs to a time-space domain reference background; carrying out nonlinear transformation on the conditional probability through negative logarithm checking so as to extract time-space conditional information; carrying out weighted summation on the conditional information of image in an adjacent domain through considering the local consistency of image characteristics; and as characteristics, carrying out object detection by using a linear classifier. The conditional probability is rapidly calculated by using a color histogram, and an image block replacing a single pixel is adopted for carrying out modeling and detection, thereby reducing the algorithm complexity and the storage space requirements; and through combining with an image block difference pre-detection mechanism, the object detection speed is increased. The method disclosed by the invention is low in algorithm complexity, less in storage space requirements and high in algorithm instantaneity, and can effectively suppress the background disturbance interference and isolate the noise influence; and by using the method, the real-time detection of moving objects on the existing computers is realized, therefore, the method is applicable to embedded intelligent camera platforms.

The invention relates to a dynamic self-adapting wireless sensor network invasion detection intelligence algorithm and belongs to the wireless sensor network information safety technology field. The algorithm comprises the following steps of using a min-max standardization method to carry out normalization on a network characteristic; through a mean value drift algorithm, clustering training data into a plurality of clusters, and merging into two clusters according to a relative distance among cluster centers; taking normal data as a template and marking the two clusters as normal clusters or abnormal clusters; distributing weights for each characteristic vector of the training data according to a distance between the characteristic vector and a cluster center where the characteristic vector is located; taking the marked and weighted training data as input of a weighted support vector machine so as to construct a decision function; through the decision function, determining whether the test data is normal or abnormal; and during a detection phase, adding detection data which is determined by the decision function into the training data at intervals of updating time so as to reconstruct the decision function. The algorithm and disposition are simple, cost is low, and the algorithm can adapt to different network structures, can detect attack behaviors with different forms and possesses an expansion capability.

The invention discloses an apple grading method and system based on machine vision, and the method comprises the steps: collecting a set number of sample images, and building a sample image set; collecting a to-be-detected sample image, and carrying out image preprocessing; performing feature extraction on the sample to be detected; respectively extracting a color feature, a fruit shape feature, a fruit diameter feature and a surface defect feature of the to-be-detected sample image; establishing a classification standard, and establishing a classification model of sample characteristics and grades by utilizing a genetic algorithm and a BP neural network; and inputting the extracted color feature, fruit shape feature, fruit diameter feature and surface defect feature of the to-be-detected sample image into the classification model, and outputting the grade classification of the to-be-detected sample. The method has the advantages that clear apple surface images can be obtained, and classification errors are reduced. The BP neural network model optimized based on the genetic algorithm can accurately classify apples, and the classification efficiency of the apples is greatly improved.

The invention provides an image classification method and a device. The method includes the following steps: extracting classification image characteristics of a to-be-classified image; quantizing each classification image characteristic into a plurality of vision words in a vision dictionary according to a similarity relation of the each classification image characteristic and the vision words in the preset vision dictionary, and determining similarity factors of the each classification image characteristic and the quantized vision words respectively; determining weight of the vision words to set a classified vision word histogram of the to-be-classified image according to the similarity factors of each vision word corresponding to different classification image characteristics in the vision dictionary; inputting the classified vision word histogram into an image classifier which is generated by training of a sample vision word histogram which is preset according to a large amount of sample images, and determining classes of the to-be-classified image according to output results. The image classification method and the device can improve accuracy of image classification and reduce classification errors.

The invention provides a zero sample image classification method based on a relative attribute random forest. An attribute ranking score model is built for an image with an unknown class according to the relative relationship between image classes and image attributes, the attribute ranking score models for all images serve as training samples to train a random forest classifier, and finally, according to an attribute ranking score for a tested image and the random forest classifier obtained through training, the label of the tested image is predicted. The method of the invention can realize zero sample image classification and has the advantages of high classification recognition rate, strong model stability and the like.

The invention discloses a hyperspectral image classification method based on a spatial-spectral convolution kernel, a storage medium and equipment. The method comprises the following steps: inputtinga three-dimensional cubic hyperspectral image; dividing a hyperspectral image data set; performing SLIC superpixel segmentation on the hyperspectral image data, and extracting a spatial convolution kernel; extracting a spectral convolution kernel; performing multiple convolution operations on the spatial convolution kernel and the spectral convolution kernel to form a spatial-spectral convolutionkernel pre-training network; and inputting the output feature map of the pre-trained network into a supervised training network for training, obtaining a cascaded supervised training network after thetraining is completed, and realizing image classification of network output features through a softmax function. According to the method, the characteristics of the images can be more effectively utilized, and rapid and high-accuracy classification of the hyperspectral images is realized.

The invention discloses a classification mechanical arm for a plastic bottle production line. The classification mechanical arm comprises a classification box; a conveying cavity is formed in the classification box in a left-right penetrating mode; a conveying belt and a separation belt are slidingly arranged in the conveying cavity; the conveying belt is positioned on the front side of the separation belt; conveying devices are connected into the conveying belt and the separation belt; and the conveying devices can drive the conveying belt and the separation belt to slide. The classificationmechanical arm can automatically detect the diameters of bottle bodies of all plastic bottles to improve the detecting efficiency, and can perform classification according to the diameters of the bottle bodies of the plastic bottles; and the sizes of the plastic bottles can be reflected through the diameters of the bottle bodies more accurately, so that the classification errors are effectively reduced.

The invention belongs to the technical field of remote sensing image recognition, and discloses a beet recognition method based on single-time-sequence NDVI, and the method comprises the steps: obtaining near-infrared band, red light band and remote sensing image data; screening spectral characteristics of crops, finding out a period with the most obvious beet spectral characteristic value, and carrying out single-time-sequence NDVI inversion analysis; carrying out statistical analysis on altitude and slope information of all beet plots; taking single-time-sequence remote sensing data as basicdata, importing the obtained GPS coordinate information into an image, performing classification by using a supervision classification method, and identifying beet in a research area by using a random forest classifier to form an identification result graph; performing cumulative statistics on the single-time-sequence NDVI image data, and determining a lower limit threshold value and an upper limit threshold value of crop classification; and taking the altitude, the gradient and the preliminary classification threshold of the random forest classifier as screening conditions through a classification tree method. According to the invention, the beet recognition rate is improved to the greatest extent.

The invention discloses a torch smoke dust monitoring method based on visual perception, and belongs to the crossing field of image processing and environment perception. According to the VMFM provided by the invention, firstly, whether flame exists in an image is identified by utilizing a wide tuning color channel, then the position of the flame is determined by combining rapid significance detection and a K-means method, finally, a potential torch smoke area is searched by taking the flame area as a center, and finally, whether torch smoke exists or not is detected. Experimental results on aplurality of video sequences acquired from a petrochemical plant show that the method is superior to existing related methods in monitoring performance and calculation efficiency. According to the torch smoke dust monitoring method based on vision, torch smoke dust can be found in time, and sufficient combustion of torch gas is guaranteed.

The invention relates to a design method of an elastic network constraint self-interpretation sparse representation classifier. The method comprises the following steps: training samples are read, the training samples are linearly transformed to a high-dimensional kernel space, each type of the training samples are learnt in the high-dimensional space, a contribution (i.e., a weight) made by each individual in the type of the training sample to constructing a sub-space of the type of the training samples is found, and a dictionary is constructed by a product of the type of the training samples and a weight matrix; and elastic network coefficient coding of test samples in the kernel space is obtained through training the obtained sparse representation dictionaries, and finally, the test samples are fitted by use of each type of the dictionaries and the elastic network sparse coding corresponding to the dictionaries, fitting errors are calculated, and the type of minimum fitting errors are the type of the test samples. According to the invention, the method is integrated with the advantages of ridge regression and lasso regression, sparse coding features of the samples are enabled to sparse, the fitting errors are also quite small, classification errors are effectively reduced, and the identification performance of a classifier is improved.

The invention discloses a video classification method based on a recurrent neural network. On the basis of existing GRU video classification, in the GRU training process, Taylor series are innovatively used for explaining a gating cycle unit, and parameter training of the GRU is assisted by introducing remainder items in the Taylor series in the GRU training process. In the Taylor series, the error is related to the order of the series, and the larger the order is, the smaller the error is, so that the video classification error is reduced by constructing the high-order Taylor series for assisting the GRU training. Meanwhile, the relation between the gating circulation unit and the Taylor series is established, and further, the generalization ability of the GRU is explained through the relation.

The invention discloses a deep cluster neural network model construction method for aerodynamic data processing, and relates to the technical field of deep cluster neural network model construction in the aerodynamic data processing field. The method comprises the following steps: a, preparing and preprocessing a data set: firstly, obtaining an aerodynamic data set through a computational fluid dynamics method, and extracting main design parameters and response parameters in the aerodynamic data set; classifying the data, and dividing the data into a plurality of subsets; labelling each subset ; finally, dividing the pneumatic data set into a training set, a verification set and a test set; b, constructing a deep cluster neural network model; c, training a deep cluster neural network model; and d, verifying the deep cluster neural network model. When the cluster neural network obtained through training of the method is used for processing aerodynamic data, the defect of a traditional neural network model in the environment of insufficient sample quality caused by uneven sample sampling or uneven sample distribution can be overcome, and therefore the prediction precision of the model is improved.

The invention provides a high-voltage cable partial discharge mode recognition method, and the method comprises the following steps: obtaining training data through laboratory data, dividing the training data into a training set and a test set, and collecting onsite to-be-recognized data; respectively inputting the training set into a plurality of classifiers for training, respectively inputting the test set into the trained classifiers for testing, and inputting to-be-identified data acquired on site into the trained classifiers to obtain a preliminary identification result; taking the preliminary identification result as input data, constructing a partial discharge mode identification model by adopting a learning method taking a support vector machine as a classifier, and then inputtingthe input data into the partial discharge mode identification model for testing to obtain a tested partial discharge mode identification model; and inputting to-be-identified data acquired on site into the tested partial discharge mode identification model, and outputting to obtain a high-voltage cable partial discharge integrated mode identification result.

The invention discloses an image tracking method and device, and the method includes: an image tracking method. The tracking method includes: determining a plurality of original areas corresponding to different target images in the video; expanding the boundaries of the original areas in the video according to a predetermined ratio to obtain a plurality of target areas, so that the multiple target areas are located The range of the image is greater than the range of the image where the multiple original areas are located; the multiple target areas are scaled to a predetermined size to obtain multiple predetermined target areas. The present invention expands the window of the original tracking object by a certain proportion, and then scales it to a fixed size, thereby greatly reducing the number of detection and search windows, and aiming at the commonality of similar multiple targets, all target images share the same Learners, online models and detectors to improve the tracking efficiency of target images.

The invention discloses a classified storage method and a system based on big data, relates to the technical field of information, and is used for realizing that the big data is classified into structured data, semi-structured data and non-structured data, and the structured data are directly stored by adopting a relational database; mode extraction is performed on the semi-structured data, the semi-structured data are mapped into structured data, and the structured data are stored by adopting an object-based relational database; the non-structured data are subjected to feature extraction, virtual classification and object creation to be converted into structured data to be stored. The method has the advantages of being small in classification error, high in efficiency, good in compatibility and capable of achieving normalization and standardization of information data management.

The invention relates to a power distribution network fault classification method and system based on deep learning, and a medium. The method comprises the steps: obtaining a plurality of original fault waveform data sets; respectively processing each original fault waveform data set to obtain target sample data; making all target sample data into a data set, dividing the data set into a trainingset and a test set, constructing a deep learning network model, and training the deep learning network model by using the training set to obtain an original fault classification model; performing parameter tuning on the original fault classification model by using the test set to obtain an optimized fault classification model; acquiring a real-time fault waveform data set, processing the real-timefault waveform data set to obtain to-be-detected fault data, and performing real-time identification on the to-be-detected fault data by utilizing the optimized fault classification model to obtain areal-time fault classification result. According to the method, the faults in the power distribution network are quickly and reliably recognized and classified by utilizing the strong classificationadvantage of deep learning, the recognition efficiency is high, and the classification accuracy is high.