32results about How to "Increase the number of features" patented technology

The invention provides a human body movement recognition method based on convolutional neural network feature coding and mainly aims to solve the problems of complicated calculation and low accuracy in the prior art. According to the implementation scheme, TV-L1 is utilized to obtain a video light steam graph; convolutional neural network coding, local feature accumulation coding, dimension-reducing whitening processing and VLAD vector processing are sequentially performed in a video space direction and a light stream movement direction, and space direction VLAD vectors and movement direction VLAD vectors are acquired; and information in the video space direction and information in the light steam movement direction are merged to obtain human body movement classification data, and then classification processing is performed. According to the method, convolutional features are subjected to local feature accumulation coding, so that the recognition rate is increased when complicated background data is processed, and the calculated amount is reduced; the features acquired by fusing video VLAD vectors and light stream VLAD vectors has higher robustness to environmental changes, and the method can be used for performing detection and recognition on human body movement in a monitoring video in areas such as a community, a shopping mall and a privacy occasion.

The invention relates to the field of anti-virus of computers and provides a malicious code detection method and system. The method combines binary system feature codes and heuristic detection and comprises the following steps of: loading a virus feature base, a detection point and a detection identifier by taking an unknown virus detection method as a feature point; analyzing a detection target, and if the detection point appears, recording the identifier corresponding to the detection point; judging whether to convert and shorten the identifier; matching the processing result which is regarded as a feature with features in the loaded virus feature base; judging whether to carry out secondary detection; if so, detecting by using a next virus feature base; otherwise, reporting the result. By the adoption of the method and the system provided by the invention, the problem of low speed of the heuristic detection is solved; in addition, since a feature acquisition method and the detection method are separated, the capacity of cross-platform detection is improved.

The invention relates to a block and sparse principal feature extraction-based rapid collaborative saliency detection method. The method sequentially includes the four steps of feature extraction of image blocks and original pixels, sparse principal feature extraction, clustering-based feature computation and collaborative saliency map generation. With the block and sparse principal feature extraction-based rapid collaborative saliency detection method adopted, limitations of application of existing collaborative saliency detection methods to various kinds of aspects which are brought by problems of memory occupation and time consumption which are further caused by ignoring of other saliency targets as well as increasingly high-definition images and the increase of the number of processing images in the prior art can be avoided.

The invention discloses a steganalysis algorithm specific to H.264 video motion vector information embedment. The algorithm aims at rapidly and effectively carrying out steganalysis to an H.264 video, and especially has relatively high detection rate specific to an embedding algorithm which keeps statistic characteristics unchanged and a low embedding rate relatively well. The technical scheme used by the invention is that motion vectors are accurately extracted; high order differential characteristics are designed on the basis of first order differential characteristics; according to this kind of characteristics, steganalysis can be carried out relatively well on a steganography algorithm which compensates vector statistic characteristics; multiple simple classifiers are trained by using integrated learning thought; an integrated steganography classifier based on FLD is designed; and the integrated classifier is relatively greatly improved in classification speed; and compared with single same classifiers, the integrated classifier has higher classification accuracy.

The present invention relates to a neighborhood entropy and consistency detection-based SAR (synthetic aperture radar) image registration method. The objective of the invention is to solve the technical problem of low registration accuracy of an existing feature-based image registration method. The method includes the following steps that: a bilateral filtering algorithm is adopted to filter an SAR image reference map and an image to be registered; a BRISK (Binary Robust Invariant Scalable Key point) algorithm is adopted to perform feature extraction on the SAR image filtered reference map and the filtered image to be registered; distance matching is adopted to obtain initial matching point pairs; wrong matching point pairs are removed through using the neighborhood entropies of feature points; wrong matching point pairs are removed through using a consistency detection algorithm; an affine transformation matrix between the SAR image filtered reference map and the filtered image to be registered is calculated; the affine transformation matrix is used to transform the filtered image to be registered, so that a transformed filtered image to be registered can be obtained; and an optimal affine transformation matrix is obtained through using a genetic algorithm, affine transformation is performed on the filtered image to be registered, and the transformed filtered image to be registered and the filtered reference map are displayed in a superimposed manner.

The invention discloses a detection method for a picture small target. The method includes using a to-be-detected picture as the input of a target detection network; obtaining six feature maps with different sizes from the to-be-detected picture, performing feature fusion on the bottom-layer feature map of the pyramid and the high-layer feature map of the pyramid in the six feature maps with different sizes by adopting a bilinear interpolation method to obtain six new feature maps with different sizes; considering the relationship among the scale feature maps; the feature map participating inprediction has richer detail information and semantic information; training the target detection network by using default setting of a pre-training model, using a random gradient descent algorithm tocarry out optimization training on the loss function expression, so rapid convergence of the model is facilitated, the trained target detection network is utilized to carry out testing, to-be-detectedimage characteristics are extracted, detection precision and detection speed of the picture small target can be effectively improved, and detection precision of the small target can be effectively improved under the condition of completely satisfying real-time performance.

The invention is suitable for the technical field of image processing, and provides an image classification method, an image classification device and terminal equipment, and the method comprises thesteps: obtaining a to-be-detected image; obtaining global features of the to-be-detected image, wherein the global features refer to feature vectors extracted from the whole to-be-detected image; acquiring local features of the to-be-detected image through a random frame, wherein the local features refer to feature vectors extracted from a local area of the whole to-be-detected image; and inputting the global features and the local features into a classifier, wherein the classifier outputs the category of the to-be-detected image. According to the invention, the accuracy of image classification can be improved.

The invention provides a mutual capacitance type pressure sensor, a touch display device and a pressure sensing method. The touch display device comprises a touch display panel, a backlight module anda pressure sensor, wherein the backlight module is arranged on the back surface of the touch display panel; the pressure sensor comprises a first conduction layer, a second conduction layer, an insulation layer and a third conduction layer; the first conduction layer is arranged between the touch display panel and the backlight module; the second conduction layer is arranged between the first conduction layer and the backlight module; the insulation layer is arranged between the first conduction layer and the second conduction layer; the backlight module is arranged between the second conduction layer and the third conduction layer; and the first conduction layer, the second conduction layer and the third conduction layer are electrically insulated from each other.

The invention discloses an image recognition method and device, electronic equipment and a readable storage medium. The image recognition method comprises the steps of determining a first feature mapof a current frame image through a convolutional neural network based on the type of the current frame image; determining a second feature map of a key frame image before the current frame image; performing feature alignment on the first feature map and the second feature map to obtain a first alignment feature map; fusing the first feature map and the first alignment feature map to obtain a firstfused feature map; and based on the first fusion feature map, identifying the content in the current frame image. Thus, the features in the current frame of image can be fused on the basis of featurealignment for image recognition, the number of the features for image recognition is increased, and the accuracy of image recognition can be improved by performing image recognition on the basis of the increased number of the features.

The invention discloses an indoor positioning method based on geomagnetic fingerprint matching, and the method specifically comprises the following steps that a user terminal firstly provides an indoor environment map and indoor magnetic fingerprint information for a server, and the server generates a magnetic map through a magnetic map construction algorithm and stores the magnetic map in a server database; after the user terminal detects that the user moves, the user terminal generates a magnetic fingerprint and user movement information of the current position, initiates a positioning request to the server and delivers the magnetic fingerprint and the user movement information to the server; after the server receives the positioning request and the user data, a positioning algorithm isstarted for positioning, and a positioning result is returned to the user terminal after positioning operation is completed; and after the user terminal obtains the positioning result, the positioningresult is displayed on the user terminal interface. According to the method, the aspects of magnetic fingerprint generation, magnetic map construction and matching positioning are optimized so that the positioning precision and the positioning real-time performance are improved while the positioning cost is reduced.

The invention discloses an MB-SSD method suitable for target detection. The method comprises the steps: extracting a plurality of small target images, and carrying out the enhancement of the extractedplurality of small target images through employing a generative adversarial network; constructing an MB-SSD feature extraction network, the MB-SSD feature extraction network comprising a main branchfeature extraction network, a branch feature extraction network and a positioning network, inputting the enhanced data into the MB-SSD feature extraction network, and respectively obtaining classification positioning results of the main branch feature extraction network and the branch feature extraction network; adjusting the output characteristics of the classification network according to the IoU overlap ratio of different candidate boxes in the same area on the positioning network; fusing the classification positioning results of the main branch feature extraction network and the branch feature extraction network, and performing dimension reduction; calculating model loss, training the model, and optimizing model parameters. According to the method, a method for adjusting the classification result by adding the relative overlap ratio into the classification layer can be added to improve the classification effect of the SSD algorithm, Meanwhile, the detection precision of the small target is effectively improved.

The invention provides a lung lobe segmentation system based on a three-dimensional convolutional neural network, and the system comprises a data obtaining module which is configured to obtain to-be-recognized chest CT image data and carry out the preprocessing of the to-be-recognized chest CT image data; the lung lobe segmentation module is configured to input the preprocessed lung CT image data into a preset three-dimensional convolutional neural network model to obtain a lung lobe segmentation result; wherein the three-dimensional convolutional neural network model comprises three lower sampling layers and three upper sampling layers, the lower sampling layers are realized through convolution layers, and the upper sampling layers are realized through deconvolution layers. According to the method, segmentation of different lung lobe regions can be rapidly realized, and the method has relatively high robustness for wide clinical data and relatively high generalization for data of different hospitals.

The invention provides a target detection method and system based on a target detection model; the method comprises the following steps: obtaining an image data set, and dividing the image data set into a training set A and a verification set B; establishing a target detection model framework, wherein the target detection model framework adopts a YOLOv4 model configured with a Fish activation function in a CSPDarknet53 network module; and training the target detection model framework through the training set A, and testing the target detection model framework through the verification set B to generate a target detection model. Under the same condition, the training time is saved by 2.26% by using the Fish activation function compared with the training time by using the Mish activation function, the total loss is reduced by 4.32%, the Val loss is reduced by 6.37%, and the detection confidence is improved on the actual detection result.