71results about How to "Improve image recognition" patented technology

The invention discloses an intelligent luggage barrow of an airport. The intelligent luggage barrow is characterized in comprising a main body frame, a control panel, a deep visual identification device, a control box and a chassis driving device, wherein the main body frame comprises a luggage load bearing device and a functional frame; the functional frame comprises two supporting rods and an advertising space panel arranged between the two supporting rods; a small part luggage frame is arranged between the supporting rods on the top; two sides of the top end of each supporting rod are provided with panel supporting frames; the control panel is embedded into the panel supporting frames on two sides and is fixed by bolts; the deep visual identification device is fixed on a rear plate of the small part luggage frame; the chassis driving device is arranged on a wheel axle, and the supporting rods above the chassis driving device are riveted with the control box; a control system is integrated in the control box. According to the intelligent luggage barrow, functions including autonomous navigation, intelligent following, voice interaction, service consultation, on-line entertainment, convenient shopping, automatic weighting, forbidden article detection, automatic conversation, one-key recovery, kid missing prevention, automatic printing of boarding checks and the like are combined, and the intelligent luggage barrow of the airport can lead intelligent, fashionable, relaxed and convenient brand-new airport service experience.

The invention discloses a leaf disease identification method based on a machine learning algorithm, belonging to the technical field of image processing. The invention firstly performs preprocessing operations such as grayscale transformation, image enhancement and denoising on the collected leaf sample image; Then the preprocessed image is segmented by adaptive threshold algorithm to represent the texture information of the sample image effectively. The RGB color space is selected to extract the color features of the sample image, and the texture features of the segmented image are extractedaccording to the gray level co-occurrence matrix. Finally, support vector machine model is selected to classify and recognize the sample images by cross-validation algorithm. Firstly, the main parameters of SVM model are optimized by grid optimization method, and then the parameters with the best recognition accuracy are selected to establish SVM classification and recognition model. The inventioncan enable the computer to automatically identify the diseases and pests of the leaves through training, greatly reducing the space and time overhead, and improving the identification accuracy, and has the characteristics of fast, accurate and strong robustness.

The disclosure provides noncontact type book scanning equipment. The noncontact type book scanning equipment comprises a depth information acquisition component which acquires depth information of a book to be scanned; a color information acquisition component which acquires color information of the book to be scanned; and a server terminal which receives the depth information acquired by the depth information acquisition component and the color information acquired by the color information acquisition component. The server terminal comprises a three-dimensional modeling module which is used for constructing a three-dimensional mesh model of the book to be scanned based on the received depth information; and a two-dimensional flattening module which is used for recovering a two-dimensional flat image of the book to be scanned based on the three-dimensional mesh model of the book to be scanned and the received color information. The scanning effect can be enhanced, reduction of image resolution can be improved and user reading experience can be improved by the noncontact type book scanning equipment.

Disclosed are image recognition apparatuses and methods which detect subject objects by configuring effective candidate regions for the subject objects in a target image. The apparatus includes an image inputting part receiving at least one image; an effective candidate region configuring part configured to receive a user input including information on reference positions of effective candidate regions and an upper limit and a lower limit of the effective candidate regions, and to configure the effective candidate regions based on the user input; and an effective region determining part configured to select an effective region meeting a preconfigured threshold among the effective candidate regions configured based on the user input. Therefore, since a complex conventional camera calibration procedure is not necessary, time and cost can be saved.

The invention discloses a nonnegative matrix factorization method based on a discriminative orthogonal subspace constraint. The method mainly comprises the following steps of (1) stretching an image in a training sample set into vectors to compose a training data matrix Xtrain, then factorizing the Xtrain in a nonnegative matrix factorization framework based on the discriminative orthogonal subspace constraint, and directly exerting a discriminative constraint item based on within-class and between-class associations to the basis matrix; (2) constructing a projection matrix W by use of the learned basis matrix U*, calculating projection expression of the training data Xtrain and test data Xtest in the projection matrix W, and performing an image recognition experiment with a nearest neighbor classifier; and (3) calculating the image identification precision. According to the nonnegative matrix factorization method based on the discriminative orthogonal subspace constraint, the discriminative structure information inside the data are explored and utilized, the discriminative constraint directly exerted to the basis matrix in the algorithm enhances the generalization performance of the algorithm and improves the image identification effect; and the method can be widely applied to the field of data mining and data analysis.

The invention discloses a multi-camera high-precision pedestrian re-identification method for supervised scenes in the same camera. The method comprises the steps of photograhing by adopting multiplecameras in the same pedestrian scene, selecting a basic network model, modifying after pre-training, acquiring a pedestrian picture set to be trained, establishing pedestrian memory features for eachcamera, and initializing the pedestrian memory features; based on an existing to-be-trained pedestrian picture set, performing training optimization and supervision in the same camera stage on the basic network model; obtaining a pedestrian pseudo-label by combining the trained pedestrian memory features with a clustering method, and performing fine tuning training on the basic network model by using the pedestrian pseudo-label; and performing cross-camera pedestrian re-identification application on the basic network model obtained by training. According to the method, the recognition performance can be effectively improved only by marking the pictures in the same camera in the scene, the re-recognition accuracy equivalent to that in a fully-supervised scene is achieved, and the pedestrianre-recognition accuracy equivalent to that in the fully-supervised scene is achieved.

The invention aims at providing a method of target identification and an apparatus thereof. The method comprises the following steps of acquiring a plurality of identification results for an identification target; based on the plurality of identification results, determining a usable result in the plurality of identification results. According to a scheme of the invention, through screening the plurality of identification results in image information, possible useless identification results are removed so that useless identification data storage is reduced, a processing capability of user equipment is increased, subsequent identification operations, such as face searching and the like, can be well performed and an identification effect to the image information is increased.

This application provides a method for constructing a network structure optimizer performed by an electronic device. The method includes: performing feature extraction on a network structure of an image recognition neural network by using a network structure optimizer, to obtain feature information corresponding to the network structure; predicting the feature information by using the network structure optimizer, to determine a plurality of optimization manners for the network structure; updating the network structure of the image recognition neural network according to the optimization manners for the network structure, to obtain an optimized network structure of the image recognition neural network; and determining an image recognition performance parameter of the optimized network structure of the image recognition neural network, and updating a parameter of the network structure optimizer according to the image recognition performance parameter, the network structure optimizer being configured to optimize the network structure of the image recognition neural network.

The invention provides a method for detecting a soil-wetted body in a drip irrigation mode. The method comprises the following steps of collecting and processing GPR data, measuring the soil volume water content, establishing a relation model between the soil volume water content and the dielectric constant, and interpreting the processed GPR data. According to the invention, the established relation model between the soil volume water content and the dielectric constant is more accurate compared with a traditional Ttop model. The method provided by the invention is used for carrying out original detection on the premise that the soil-wetted body is not damaged at all. The change of the time scale and the spatial scale of the soil-wetted body is fully considered, so that the soil-wetted body can be rapidly and efficiently detected in the nondestructive and in-situ manner based on the ground penetrating radar technology. The application of the ground penetrating radar technology in thesoil-wetted body monitoring field is promoted. Meanwhile, the technical support is provided for the accurate agriculture and the water-saving agriculture.

The invention relates to a video processing method and device, a computer readable storage medium and computer equipment. The method comprises the following steps of acquiring a to-be-processed video;segmenting at least one target video from the to-be-processed video; acquiring background game data corresponding to the target video from a game operator background, wherein the background game datahave a corresponding game; extracting a key frame from video frames of the target video; in game frames of the game, determining a target game frame corresponding to the key frame; establishing a mapping relationship by adopting the target game frame and the key frame; and according to the mapping relationship, aligning the target video and the game. When a game commentary is required to be embedded in a certain position of the game, the embedded position in the video can be determined according to the mapping relationship, so that precise embedding of the game commentary is realized.

An image capture apparatus including a light guide element, an image capture device and a light emitting device. The light guide element has a first side, a second side and a light emitting portion located at the second side. The light emitting portion includes a plurality of enhanced transmission microstructures. The image capture device is disposed on the second side of the light guide element corresponding to the position of the enhanced transmission microstructures. A light beam, which is generated by the light emitting device and transmitted at least by the light guide element, is totally reflected to form a signal light beam. Thereafter, the signal light beam passes through the enhanced transmission microstructures and then is received by the image capture device.

An image display apparatus includes a frame along a head of an observer, a display unit that displays an image that can be visually recognized by the observer, and a supporting part attached to the frame and supporting the display unit, wherein the frame is fixed to a position in which the image displayed by the display unit connected to the frame via the supporting part is visually recognized by the observer.

The invention discloses a feature fusion method. The method comprises the steps of obtaining a plurality of images of a target object; inputting the plurality of images into a pre-trained image quality model to obtain the M-dimensional image quality of each image; inputting the plurality of images into a pre-trained feature recognition model to obtain N-dimensional image features of each image; performing weighted summation on the M-dimensional image quality and the N-dimensional image characteristics to obtain fusion characteristics, wherein M is a positive integer, N is a positive integer, when M is equal to 1, N is greater than or equal to 2, and when M is greater than or equal to 2, M is equal to N. The invention further provides a feature fusion device, an electronic device and a storage medium. According to the present invention, the image recognition effect can be improved.

The invention discloses a supervised image classification method based on a self-paced constraint mechanism. The method comprises the following steps: dividing difficult types of training samples; establishing a sparse representation model, and substituting samples into the sparse representation model for training; obtaining an image classification model and a prediction model; constructing a category decision-making device, wherein the training sample difficult type comprises a training easy sample and a training difficult sample, and the division training sample difficultly-easy types are divided by adopting a self-step constraint matrix. According to the invention, the training samples are divided through the self-paced constraint matrix; the easy training sample and the difficult training sample are sequentially substituted into a defined sparse representation model for continuous training; a specific self-step constrained image classification scheme can be formed, more judgment information can be conveniently utilized, robustness is achieved on sample noise, and therefore the problem that a supervised dictionary learning mechanism is not suitable for complex samples containingnoise and huge intra-class changes can be solved, and the image recognition effect is improved.

A system and an imaging method for using photoacoustic effect are provided in the present invention. The system includes a light source for generating a light beam, a wave-guide probe and an ultrasound receiving device. The wave-guide probe further has a reception portion and at least one transmission portion. The reception portion receives the light beam and then triggers a photoacoustic effect inside the reception portion so as thereby to generate at least one sound wave thereinside to be further transmitted to the at least one transmission portion. The transmission portion is merged into the organic medium. When the sound wave is transmitted to the transmission portion, an ultrasound area is generated inside the organic medium. The ultrasound receiving device is located adjacent to the organic medium, receives the ultrasound generated in the ultrasound area to form an ultrasound image of the organic medium, so as to achieve the imaging method.

The invention relates to an image super-resolution reconstruction method and device, a storage medium and computer equipment, and the method comprises the steps: carrying out the downsampling of a high-resolution image sample, obtaining a low-resolution image sample, and extracting the low-resolution image features of the low-resolution image sample through a first convolution module of a machinelearning model; enabling the second convolution module to perform multiple times of nonlinear mapping on the low-resolution image features to obtain high-resolution image features; constructing a high-resolution reference image according to the high-resolution image features; calculating a loss value between the high-resolution reference image and the high-resolution image sample; performing backpropagation on the loss value in a machine learning model to adjust model parameters of the machine learning model according to the loss value; acquiring the low-resolution image; and reconstructing the low-resolution image into the high-resolution image by using the machine learning model obtained after training. The problem of poor recognition effect caused by low resolution of the face image can be avoided.

A semiconductor device has a plurality of first opening portions formed in an interlayer insulating film. The surface is covered with a metal film with a surface having concavities and convexities which scatter reflected light. Size of the first opening portion is of the same level as a contact hole of a component and cannot be recognized by an image recognition apparatus. The metal film can be recognized by the image recognition apparatus. By forming a TiN film serving as a reflection prevention film on an end of the metal film, portions that can easily scatter light and a portion that cannot easily reflect light are adjacent in an alignment marker. A passivation film is formed on the interlayer insulating film and the TiN film. Recessed portions disposed in the metal film are exposed to a second opening portion formed in the passivation film and the TiN film.

An object of the present invention is to maintain the ease with which a template is created based on design data without acquiring an actual image, which is achieved by providing the template with equivalent information contained by a template used for image recognition that involves same-type image comparison, and to improve image recognition performance by increasing the matching rate between a template and an actual image.To achieve the above object, the present invention provides a method, apparatus, and program for creating based on design data a template that is used for image recognition, wherein luminance information is set for each area in the template based on the material information of the region defined by the template. Specifically, the luminance information is set based on at least one of information from among the above material information, the pattern size information of a pattern arranged in the region defined by the template, the setup conditions of an imaging apparatus, the layer information of the region defined by the template, and the outline information of a pattern.

To create a template for use in image recognition based on design data, luminance information is set for each area in the template based on the information regarding the region defined by the template. The luminance information may be set based on material information, pattern size information of a pattern arranged in the region defined by the template, and layer information of the region defined by the template. Alternatively, luminance information may be set based on material information, setup conditions of the scanning electron microscope, and pattern outline information of a pattern arranged in the region defined by the template.

A microelectromechanical probe is manufactured by a MEMS manufacturing process forming a probe body and a cutting process providing a pinpoint portion a cutting face. The probe has a top surface, a body portion, and a pinpoint portion extended in a probing direction from the body portion and provided with first and second sides and a probing end oriented in the probing direction. The cutting face is provided on the top surface, adjoins the first and second sides and the probing end, and has at least one cut mark formed by the cutting process, extended from the first side to the second side and non-parallel to the probing direction. The cutting face descends from an edge cut mark to the probing end.

The embodiment of the invention provides a semi-supervised semantic segmentation model training method, a semi-supervised semantic segmentation model identification method and a semi-supervised semantic segmentation model identification device. The semi-supervised semantic segmentation model training method of the embodiment comprising obtaining first supervised data obtained by manually annotating a to-be-annotated object in a first image, and then obtaining a full-supervised semantic segmentation model with a relatively high recognition rate of the to-be-annotated object through training ofthe first supervised data; labeling the to-be-labeled object in the second image which is not manually labeled by utilizing the fully-supervised semantic segmentation model to obtain second superviseddata; training a semi-supervised semantic segmentation model by utilizing the first supervised data obtained through manual annotation and the second supervised data obtained through full-supervisedsemantic segmentation model annotation, and identifying the first image, the second image and the random disturbance item by utilizing the semi-supervised semantic segmentation model to obtain third supervised data; and finally, training the semi-supervised semantic segmentation model again through the first supervised data, the second supervised data and the third supervised data.

The embodiment of the invention provides an image detection method, apparatus and device, and a computer storage medium. The method comprises the steps of obtaining to-be-detected image data of a to-be-detected image in a grayscale color space; determining a brightness accumulation histogram of the to-be-detected image data in the gray scale color space; determining the slope of the brightness accumulation histogram variation of each predetermined brightness interval relative to the brightness value; comparing the corresponding slope of each predetermined brightness interval with a preset slope threshold value; and determining that the to-be-detected image is related to a backlight scene if it is counted that the number of the predetermined brightness intervals of which the corresponding slopes are greater than a preset slope threshold meets a preset requirement. According to at least one embodiment of the invention, the image recognition performance under the condition of local backlight can be improved.

The invention discloses a terminal and a fingerprint imaging method. The terminal comprises a collection module used for collecting a fingerprint, an image processing module used for preprocessing the image of the fingerprint to generate a grayscale map of the fingerprint, a statistical module used for doing statistics of the grayscale distribution in the grayscale map and generating a grayscale distribution map, a processing module used for cutting a section with the gray values smaller than a first threshold and / or a section with the gray values greater than a second threshold from the grayscale distribution map and stretching the grayscale of the grayscale distribution map after cutting according to preset rules, and an output module used for taking a fingerprint image corresponding to the grayscale distribution map after stretching as a target fingerprint image and outputting the target fingerprint image. Through the technical scheme, fingerprint imaging data of different skin colors can be processed differently, the details of user fingerprints can be reflected more accurately by enlarging difference, the ability of image identification is improved, and the user experience is improved.

The embodiment of the invention provides a device, system and method for drawing a landmark map based on binocular vision. The device comprises a first lens, a first prism, a second prism, an image sensor, an image processing module and a GNSS differential positioning module. The first lens and the first prism are symmetrically arranged along the central axis of the second prism, wherein the first lens and the first prism located on one side of the second prism sequentially form a left light path, and the first lens and the first prism located on the other side of the second prism sequentially form a right light path. The light along the left light path and the light along the right light path are reflected to the image sensor through the second prism. The image sensor is used for obtaining an image collected by a binocular lens and transmitting the image to the image processing module. The image processing module is used for analyzing and processing image data. The GNSS differential positioning module is used for obtaining the coordinate of the original point of the lens coordinate system in the world coordinate system through GNSS differential positioning. The requirement on a processor is reduced, the precision is high, and the cost is low.

An object detection apparatus mounted in a system for detecting a target object in various changing environmental conditions. In the apparatus, an acquirer acquires either or both of information indicative of an external environment around the system and information indicative of an operating state of the system. A determiner determines an environmental condition with respect to an input image according to the acquired information. A setter sets, for each of plural image recognition methods each being a combination of one of the plural image recognition dictionaries and one of the plural image recognition techniques, a weighting factor according to the environmental condition determined by the determiner. A detector detects the object in the input image by applying each of the plural image recognition methods to the input image to obtain recognition results reflecting the weighting factors, and collectively evaluating the recognition results.

An image processing method of a transparent display system includes providing the transparent display system, where the transparent display system includes an image processor and a transparent display panel. The image processor receives an input image. The image processor enhances the chroma of the image according to a predetermined weighting value to produce an enhanced chroma image. The image processor sharpens the contour of the image to produce a sharpened image data. The image processor adds the enhanced chroma image to the sharpened image data to produce an adjusted image. The transparent display panel displays the adjusted image.

The invention belongs to the technical field of image processing and discloses a natural mode-based Logo generation method and an intelligent Logo generator. Source codes are written according to control rules and a core algorithm, and intelligent Logo design output is realized. The intelligent Logo generator comprises a conversion module, a form element representation module, a color representation module, a font representation module, a rule conversion module, a parameter association module, a font interface module and a logic relationship module. By use of design demands (generally from customers and competitive markets), a natural mode matched with the design demands is searched for; and the parametric intelligent Logo generator is realized by use of the core algorithm, the source codes are written according to the control rules and the core algorithm, the intelligent Logo design output is realized, Logo forms output by the generator are ever-changing, dynamic Logo design can be developed, and dynamic propagation can be realized for improving the enterprise image identification degree.

The invention relates to an image recognition method based on an optimized deep neural network model, and belongs to the technical field of image recognition. The method comprises the following steps:S1, selecting a data set of an image as a training sample, dividing the data set into a plurality of sample sets according to batch setting of a batch number, and inputting a current sample set intoa specific deep neural network model to obtain a loss function value set corresponding to the current sample set, wherein the specific deep neural network model is used as a selected deep neural network; s2, updating network parameters of the selected deep neural network by discarding or forcibly zeroing training samples corresponding to a part of sets with smaller numerical values in the loss function value set; s3, iterating S1 and S2 for multiple times to obtain an optimized deep neural network model; and S4, inputting a to-be-identified image into the optimized deep neural network model toobtain an identified image. According to the method, the problem that the training performance of the deep neural network model under large-batch image data sets is greatly reduced is solved.