425results about How to "Extract comprehensive" patented technology

The invention provides a pedestrian re-identification method based on a CNN and a convolutional LSTM network, and belongs to the technical field of image processing. Firstly space information of codes in frames are extracted by using a group of CNN, then a frame level depth space-time appearance descriptor is obtained by using a coding-decoding frame formed by the convolutional LSTM, and finally Fisher vector coding is used so that the descriptor is enabled to describe the characteristics of the video level. With application of the mode, the characteristic representation can be extracted and the videos are enabled to be arranged sequences by the characteristic representation, and the space information is maintained and an accurate model is established.

The invention discloses an attention mechanism relationship comparison network model method. An attention relationship comparison network model for small sample learning under a small amount of labeled sample data is constructed. Based on a relational network architecture, the model is divided into a feature coding part, a feature combination part and a relational coding part, the feature coding module is used for extracting image feature information, and the feature combination part is used for recombining the extracted query image feature information with the training image feature information of each group to form a new combined feature map. The relation encoding module performs nonlinear metric learning of the network; by introducing an attention mechanism and a spectrum normalizationmethod into an end-to-end deep convolutional neural network model, the model has higher classification accuracy under the condition of small sample learning, the stability of a final training result of the model is improved, and the image classification accuracy of an existing model in small sample learning is improved.

The invention relates to an aerial image insulator real-time detection method based on deep learning. The aerial image insulator real-time detection method based on deep learning includes the steps: handing a task of extracting characteristics to a deep convolutional neural network, extracting the deep characteristic information which is more comprehensive and can preferably describe an insulator,and inputting the deep characteristic information into a detector to perform prediction inference to obtain a detection result. For the aerial image insulator real-time detection method based on deeplearning, the whole process is an end-to-end quick detection channel; a target frame is obtained after the image is input; the efficiency of subsequent automatic fault diagnosis is improved; and theaerial image insulator real-time detection method based on deep learning is conductive to reducing the retrieval pressure and intensity when the line patrol staff retrieves the mass line patrol data at present. And at the same time, the aerial image insulator real-time detection method based on deep learning also utilizes the idea of transfer learning to transfer the knowledge obtained from the past task to the current target task, so as to enable the trained model to have inheritability; whenever new data is added into an image library, the target model can continue to train new data on the basis of a source model, so as to quickly achieve the expected effect and enable the old version of model not to be of no use at all because of updating of data; and the detection model can become moreand more powerful following increase of data as time goes on.

The invention discloses a two-channel convolutional neural network-based single image super-resolution calculation method. The method comprises the steps of (1) performing fuzzy degradation processing on a known high-resolution image to obtain a low-resolution image with the same size; (2) decomposing the obtained low-resolution image after the fuzzy processing in the step (1) into a texture part and a smooth structure part of the low-resolution image, and obtaining a texture part and a smooth structure part of the high-resolution image; (3) combining the low-resolution texture part obtained in the step (2) and an original low-resolution image to obtain a two-channel input, and obtaining an output of the high-resolution texture part; (4) combining the obtained output of the high-resolution texture part in the step (3) and the original low-resolution image to obtain a final image super-resolution reconstruction result, thereby finishing super-resolution reconstruction; and (5) calculating a difference value between the high-resolution texture parts obtained in the steps (4) and (2) to obtain texture part loss, and minimizing a sum of the texture loss and image loss to optimize network structure parameters.

The invention discloses an Alzheimer's disease classification method based on a depth neural network and multi-mode images, and the method comprises the steps: taking the 2D and 3D forms of the images of N classes of Alzheimer's diseases as the input layer, taking the probability of a classification result as the output layer, and constructing a depth neural network; carrying out the preprocessing of the images of N known classes, and obtaining a training sample; training the depth neural network through the training sample, optimizing the weight of network connection, and obtaining a final depth neural network; inputting the preprocessed to-be-classified image into the final depth neural network, and outputting a classification result. In order to make the most of the feature information of the brain of an AD patient, the method introduces the multi-mode image information: an MRI image, a PET image and a DTI image on the basis of a conventional single-mode medical image classification method, and the method achieves the fusion of non-image feature CSF information and gene information.

The invention discloses an image description method and system based on a vision and semantic attention combined strategy. The steps include utilizing a convolutional neural network (CNN) to extract image features from an image whose image description is to be generated; utilizing a visual attention model of the image to process the image features, feeding the image features processed by the visual attention model to a first LSTM network to generate words, then utilizing a semantic attention model to process the generated words and predefined labels to obtain semantic information, then utilizing a second LSTM network to process semantics to obtain words generated by the semantic attention model, repeating the abovementioned steps, and finally performing series combination on all the obtained words to generate image description. The method provided by the invention not only utilizes a summary of the input image, but also enriches information in the aspects of vision and semantics, and enables a generated sentence to reflect content of the image more truly.

The invention discloses a method for partial reference evaluation of wireless videos based on space-time domain feature extraction, which relates to a method for evaluating video quality. On the basis of performing profound understanding and detailed analysis on the existing video objective quality evaluation model, by combining visual characteristics of human eyes, the invention provides the method for the partial reference evaluation of the wireless videos based on the space-time domain feature extraction to improve the conventional SSIM model. The method takes the fluency of the time domain and the structural similarity and the definition of the space domain of the videos as main evaluation indexes, and under the condition of ensuring the evaluation accuracy, the method extracts characteristic parameters of ST domains (the space domain and the time domain), establishes a new evaluation model, reduces the reference data needed by the evaluation and lowers the computational complexity to ensure that the evaluation model is suitable for evaluating the quality of wireless transmission videos in real time.

The invention discloses an image super-resolution method based on a multi-scale attention convolutional neural network. By using a multi-scale structure and an attention mechanism, the convolutional neural network can extract rich information in an image, and importance distinguishing can be performed on extracted features, so that important information and interference information can be distinguished in learning of the convolutional neural network, and the learning efficiency is improved. A dense connection mode and a residual connection mode are used, so that the back propagation of the gradient is enhanced, the problem of instability of the gradient is avoided, the reusability of the features is enhanced, the features in the low-resolution image are fully utilized, the training speed of the model is increased, and the parameters of the model are further reduced. Besides, a sub-pixel convolution layer method is used in an image amplification stage, learned pixel values are filled into a high-resolution image according to a certain rule without calculation, so that the operation is concentrated in a small-scale stage, the operation amount is reduced, and the image reconstructionspeed is improved.

The invention discloses a behavior identification method based on multi feature fusion. The behavior identification method comprises the steps that S1, preprocessing is carried out on videos of a training set and a test set; S2, feature point detection is carried out on the videos which are preprocessed in the step of S1; S3, descriptors which represent different information is extracted from feature cubes in the videos, so as to form comprehensive descriptors; S4, the descriptors, which are extracted in the step of S3, of the training set are used, and a K-SVD algorithm is used to train a dictionary; and S5, feature fusion is carried out on the descriptors, which are extracted in the step of S3, of the test set, and classification is carried out through a cascade dictionary classification algorithm. According to the invention, multi feature fusion is carried out on energy information, spatial information and time information; essential motion features are abstracted; comprehensive descriptor information extracting is realized; and the system robustness is great.

The invention discloses a traffic flow prediction method based on deep neural network integration. The method comprises the following steps: obtaining original traffic flow data, carrying out data preprocessing, constructing sample data, dividing the sample data into a training set and a test set, and enabling the sample data to be one-dimensional time series data consisting of a plurality of traffic flow measurement values; constructing a convolutional neural network prediction model for traffic flow prediction, inputting sample data into the model, training by using a gradient optimization algorithm, and calculating a variance of a prediction error for the trained model on the training set; using the convolutional neural network prediction model as an individual learner according to thevariance of the prediction error, and constructing a convolutional neural network integration model for traffic flow prediction; and predicting the traffic flow data to be tested by using the convolutional neural network integration model. Based on a convolutional neural network model, an integrated learning method is utilized, an improved traffic flow prediction method is provided, and the prediction accuracy is improved.

The invention discloses a method for predicting remaining service life of a lithium battery based on FPCA (functional principal component analysis) and Bayesian updating. The method comprises the following steps of utilizing FPCA to build a lithium battery non-parameter degrading model, predicting the remaining service life of the lithium battery based on the degrading model, and analyzing the influence of different amount of modeling data on the predicting accuracy. The method has the advantages that an empirical Bayesian method is provided for updating the lithium battery degrading model in real time; the lithium battery degrading model is corrected in real time by the Bayesian method, so as to obtain the more accurate lithium battery degrading model; the distributing of the remaining service life of the lithium battery is calculated by a parameter bootstrap method, and the confidence interval is calculated.

Provided is an inversion method for the biophysical property of a forest based on a LiDAR comprehensive waveform model. Crown segmentation is carried out, waveform information within the individual tree range is gathered and then gathered on a standing forest scale, and thus transformation of the inversion scale is more flexible and reasonable; the ground return pulse is determined by means of a digital landform and a height error threshold, so that the determination effect of return pulses of a crown layer is indirectly improved; and characteristic information of comprehensive waveforms is extracted in multiple dimensions, so that rich forest crown layer information is extracted from the waveform data in a deepened manner. The inversion method improves the precision in estimating the biophysical property of the forest, provides theoretical and method supports for explanation of the mechanism, and effectively avoids the defect that value of the waveform data is greatly reduced due to abandonment of rich continuous vertically-distributed information of the forest recorded in original waveform data.

The invention discloses a feature vector-based fast and high-precision robustness matching method. Due to the changes of the environment, the influences of the target motion and the defects of sensors, the shot images are not only influenced by the noise but also have severe grayscale distortion and geometric distortion, therefore, how to achieve and realize high precision, high matching accuracy rate, fast speed, strong robustness and strong anti-interference performance becomes a goal pursued by the matching method. The invention discloses a sub-pixel-level fast and high-precision robustness matching, comprising the steps of: respectively extracting the SIFT (Scale Invariant Feature Transform) feature vectors of two pictures to be processed; carrying out PCA (Principal Components Analysis) dimension reduction treatment on the two pictures; matching the two feature vectors by utilizing Kd-tree; then screening the obtained matching points by utilizing an RANSAC algorithm; and causing the matching to achieve the sub-pixel level through a surface fitting technology so as to obtain the feature point pair of the high-precision robustness. Furthermore, the matching speed is increased by using an even point getting method. Through the invention, the obtained image matching result has extremely high matching precision and faster running speed and the experiment effect is excellent.

The invention discloses a lithium ion battery health state estimation method based on a codec model, which comprises the following steps of: (1) acquiring battery charging and discharging period dataincluding terminal voltage and current data and the maximum discharging capacity in each charging and discharging period; and (2) constructing a codec model of an attention mechanism according to thecharacteristics of the acquired data, wherein the codec model comprises an encoder and a decoder, voltage and current values are used as encoder input, an SOH estimated value of the battery is obtained and used as decoder output, and the number of nodes of each layer is determined; (3) preprocessing and normalizing the data acquired in the step (1), inputting the data into the codec model with randomly initialized weight, and minimizing the output error of the codec model through Adam algorithm training; and (4) inputting a new test sample into the codec model trained in the step (3), and calculating a prediction error so as to evaluate the accuracy of model prediction.

The invention relates to a system and method for text classification. The system comprises the steps that an initialization module reads a text, vectorization is conducted to sentences in the text, and a two-dimensional matrix vector is generated; a first extraction module conducts convolution and pool processing to the two-dimensional matrix vector, and multiple first matrix vectors are generated; the second extraction module conducts dot multiplication between the multiple first matrix vectors and an attention matrix respectively, and multiple second matrix vectors are generated; a comprehensive representation module conducts convolution to each matrix vector, so each second matrix vector is correspondingly converted into a first-dimensional vector matrix; and a classification module inputs the multiple first-dimensional vector matrixes into Fully Contact Layer for processing respectively, inputs output values to a softmax classifier, and the softmax classifier converts matrix values into probability distributions of corresponding classes, so that the text classification is completed. According to the invention, only a few of parameters are used; a network model can be converged rapidly; representation information of a text depth is extracted; and thus, accuracy for the text classification can be increased.

The invention relates to a method for segmenting a liver and a focus thereof in a medical image, and the method comprises the steps: firstly carrying out the screening and integration preprocessing ofabdominal CT image data, dividing the abdominal CT image data into a plurality of data sets with different purposes, building a new neural network, and carrying out the initial training through employing small image data; then, storing the trained model, carrying out secondary training by using the original image and a new data enhancement mode, carrying out expansion and corrosion processing onthe predicted image, and evaluating by using a medical evaluation index; through the model prediction results trained by DL, GDL and TL loss functions, adding and averaging the prediction results of the three loss models to form a fusion feature; finally, modifying the network, wherein the three loss models are fused in a single network for training prediction. End-to-end training test can be carried out, liver and focus can be identified at the same time with high precision and high speed, doctors are effectively helped to identify CT images, time and energy consumed by doctors are greatly reduced, and the probability of misdiagnosis is reduced.

The invention discloses an urban impermeable surface extraction method based on a high-resolution satellite remote sensing image. A high-resolution satellite remote sensing image is used as a data source, and a new high-resolution urban impermeable surface comprehensive index is established by combining a normalized building index, a normalized vegetation index, an improved normalized water body index and a normalized soil index. According to the method, the difference between the impermeable surface and vegetation, water and naked bodies is considered, so that the problems of same spectrum offoreign matters and the like when the impermeable surface is extracted by a single index are avoided. Compared with a traditional method of independently using the impermeable surface index to extract the urban impermeable surface, the inversion effect of the model is good, the feature information of the impermeable surface can be mined, and a new technology and means are provided for accuratelyand comprehensively extracting the urban impermeable surface with the high spatial resolution within a large range.

The invention provides a face diagnosis method and system based on deep learning. The method includes the steps of using a large number of human face images with disease labels for training a deep learning network, carrying iteration for multiple times, after parameters updating, the deep learning network being capable of automatically and effectively extracting the facial features including eyes,ears, mouths and eyebrows, and finding the internal relation between the facial features and various diseases of the human body so as to carry out effective disease detection and screen on the face images of new subjects. The method is a non-invasive computer-aided automatic diagnosis method. According to the system, the parameters can be updated by continuously collecting the facial images of patients and training, so that the prediction judgment result is more accurate. By means of the method and the system, the problem of difficulty in diseases inspection in poverty-trapped regions can beeffectively solved, so that people can conveniently and rapidly carry out non-invasive disease automatic detection and screening through a mobile phone, computer and other terminals, and the diseasescan be treated in time. The people's living quality is improved.

The invention discloses a farmland multiple robot following land cultivation system based on a stereo visual sense visual sense and a method for the same. The farmland multiple robot following land cultivation system based on the stereo visual sense comprises a touch screen control stick type control terminal, a stereo visual sense identification positioning system, a wireless communication system, a navigation land cultivation robot and a plurality of following land cultivation robot. The touch screen control stick control terminal is used for remotely controlling the navigation robot and controlling the working states of the following robots and switching the following mode. The stereo visual sense identification positioning system comprises a plurality pairs of binocular camera, an identity identification column and an image processing module, and is used for identifying and positioning obstacles and positioning the self relative to other robots. The navigation cultivation robot and the following robot are loaded with a stereo visual identification positioning system, a robot control module and a wireless communication module. The navigation robot is remotely control to navigate and the following navigators are formed into a column, which improves the efficiency, has important meaning in the cultivation during the busy season and can be widely applied.

The invention provides a preparation process of blueberry concentrated juice with high content of anthocyanin. According to the preparation process provided by the invention, pectase is adopted for performing enzymolysis on blueberry pulp, chitosan is adopted as a clarifying agent for obtaining clear blueberry juice, a decompression concentration process is adopted for preparing the blueberry concentrated juice, and microwave sterilization is adopted for sterilizing the blueberry concentrated juice. The preparation process comprises the following steps of: selecting blueberry fruits in vaccinium ashei gardenblue species; extracting the blueberry juice, and adding the pectase to perform the enzymolysis on the blueberry pulp; clarifying the blueberry juice, and adding the chitosan as the clarifying agent for obtaining the clear blueberry juice; concentrating the blueberry juice, and adopting the decompression concentration to obtain the blueberry concentrated juice; and performing sterile filling on the blueberry juice, and adopting the microwave sterilization to sterilize the blueberry concentrated juice, wherein the content of soluble solids in the blueberry concentrated juice is 60%-75%, and the content of the anthocyanin is 1.925g/L-3.384g/L. The preparation process provided by the invention can not only effectively prevent the loss of the anthocyanin in the blueberry concentrated juice, but also shorten the concentration time, control the production cost of the preparation process and effectively protect color, luster, fragrance and other sensory qualities of the blueberry juice.

The present invention discloses a facial expression synthetic method based on rapid expression information extraction and Poisson image fusion. The method mainly solves the problem in the prior art that the expression detail information can not be extracted rapidly and effectively and can not be synthesized to a target face, and comprises the realization steps of 1) obtaining a corresponding expression template according top the face characteristic point data; 2) according to an expression shape template of a target object, deforming a neutral expression image of the target object and a non-neutral expression image of a source object to the non-neutral expression shapes of the target object; 3) extracting the deformed expression detail information of a figure expression image block in a frequency domain; 4) using a Poisson image fusion method to filter the extracted expression details of the source object; 5) using the Poisson image fusion method to synthesize the filtered expression detail information to the deformation expression of the target object to thereby obtain a final synthesis result. The facial expression synthetic method of the present invention is small in needed sample capacity and natural and vivid in synthesis result, and can be used for the figure animation rendering, the video conference and the man-machine interaction.

The invention discloses a human skeleton action recognition method, and relates to a method for identifying graphics. The method is a human skeleton action recognition method combining space-time attention and a graph convolution network. Diversity and complementarity of different feature information are fully mined; the weight value of each joint point of the space structure and the importance ofeach frame of the video sequence are adaptively adjusted by using an attention mechanism; the motion recognition of the human skeleton is carried out by using the graph convolution network, and the defects that in the prior art, space-time feature information cannot be better captured, and errors are likely to occur in the recognition of the difficult motion of the human body are overcome.

The invention belongs to the technical field of databases and aims to establish an XML document constraint model known as a bidirectional path constraint model. Based on the model, the invention discloses a new method for calculating the similarity of XML documents. The structural information of the XML documents is extracted more completely through the bidirectional path constraint of a node, so that the similarity of the XML documents is balanced more accurately. A very mature N-Gram thought in the field of natural languages is introduced and an N-Gram-based partition mode is applied in the similarity calculation of path constraint. The extraction and operation of N-Gram information are simplified by making skillful use of positive integers and weight numbers. The method can be used in the fields of XML document classification, clustering, mode extraction and the like.