67results about How to "Preserve detailed features" patented technology

The invention relates to the field of pattern recognition, computer vision and deep learning, provides a fine edge detection method based on a deep fusion correction network and a fine edge detectiondevice thereof, and aims at solving the problems that edge positioning in the image is insufficiently accurate and the detected edge is insufficiently fine. The method comprises the steps that step S1, the multi-scale features of an input image are acquired through the forward propagation part network of a convolutional neural network; step S2, the final image feature having the same resolution with that of the input image is acquired by using the method of gradually increasing the feature resolution through the reverse correction part network of the convolutional neural network; and step S3,the feature channel of the final image feature is dimensionally reduced into the single channel, and the edge detection result is generated through the fitting function. The network structure is simpler, and the acquired image feature expression further retains detail features so that the detection effect is better and the edge visualization result is finer.

The invention discloses an image processing method, an image processing device, a storage medium and a mobile terminal. The image processing method is applied to the mobile terminal and comprises thesteps of: acquiring an original image collected by the mobile terminal, wherein the color depth of the original image is at least 10 bits per pixel; determining a brightest region in the original image, and performing exposure processing on the brightest region to obtain intermediate image data; and performing logarithm conversion on the intermediate image data to obtain a target image. Accordingto the image processing method of the invention, gamma correction does not need to be carried out on the intermediate image data, but logarithm conversion is carried out, and more detail features of bright parts and dark parts of the image are reserved by using logarithm conversion; and on the other hand, exposure processing is carried out according to the brightest region, detail feature loss isavoided, and more detail features are reserved.

The invention relates to a point cloud data compacting method based on a normal included angle, and belongs to the technical field of computer three-dimensional modeling. The compacting method comprises the following steps of: (1) reading the original point cloud data; (2) acquiring a k-order neighborhood of each data point, and calculating a unit normal vector of each data point; (3) acquiring an average value V of dot products of the normal vector of each data point and normal vectors of k proximal points of the data point; (4) acquiring the curvature V' of a local region where each data point is positioned; (5) classifying all data points in a point cloud; (6) determining a sampling ratio of each class; and (7) compacting the point cloud data. Compared with the traditional method, the method has the advantages that the detail features of the original point cloud can be kept, and the time cost of complicated quadric surface fitting and curvature estimation is avoided.

The invention discloses a local curved surface change factor based scattered point cloud data compaction processing method. The local curved surface change factor based scattered point cloud data compaction processing method comprises the steps of 1 reading measured point cloud data, 2 calculating a central point of a point cloud, 3 searching dynamic K neighborhood points of the central point based on cubic grids and accordingly establishing the topological relation of scattered point cloud, 4 adopting a variance component method to calculate curved surface change factors of a k neighborhood of the central point, 5 determining the compaction rate of each cubic grid in the k neighborhood of the central point and performing even compaction in within a k neighborhood range. The topological relation of the scattered point cloud is established by establishing the dynamic K neighborhood point information of the scattered point cloud. Complicated curvature calculation is replaced by the curved surface change factors. The compaction ratio is adjusted according to the curved surface change factors Xi, even compaction within the k neighborhood range is achieved, the detail characteristic of high curvature can be protected, and planar characteristic of low curvature is also protected when the compaction degree is high. Point cloud data processing and curved surface reconstruction efficiency and accuracy are improved.

The invention relates to a point cloud reconstruction method and system based on three-dimensional point cloud data feature lightweight, high-precision mass three-dimensional point cloud data acquisition is carried out on a measured object, and a point cloud data result is closer to the real morphology of the measured object. According to the method, feature point cloud extraction is carried out on collected massive three-dimensional point cloud data, an outlier extraction method is adopted to process the surface topography of a measured object, feature point cloud data on the surface of the measured object are acquired, and a point cloud region growth segmentation method is adopted to acquire feature point cloud data at the edge of the measured object in combination with a normal includedangle criterion. On the premise of reserving the feature point cloud data, down-sampling is performed on the remaining point cloud data so that lightweight processing of the point cloud can be realized. According to the method, sliding least square fitting is carried out on the point cloud data after lightweight processing, so that the point cloud data can reserve fine features of the surface, and point cloud reconstruction is carried out on the fitted point cloud data, thereby generating a measured object entity model reserving key morphological features.

The invention discloses a dual-normal mesh model fairing method based on vertex characteristics, which mainly comprises the following steps: 1) all vertices in the mesh model are divided into characteristic points and non- characteristic points. 2) the surface normal field is constructed by using guide filter. 3) The accurate surface normal field can be obtained by filtering the normal field of the surface opposite to each surface. 4) Computing the normal directions of the vertices of the characteristic points and the non-characteristic points in the triangular mesh model respectively, so as to construct the normal fields of the vertices. 5) updating the position of the non-characteristic vertices according to the surface normal direction; The characteristic vertex positions are updated iteratively according to the surface normal and vertex normal. 6) smoothing that mesh model. The invention can better retain the detail characteristic of the mesh model while removing the noise of the mesh model, and the error of the mesh model after smoothing is small, and the mesh model can more accurately approach the actual model.

The invention relates to a characteristic-maintained point cloud data compacting method and belongs to the technical field of computer three-dimensional modeling. The compacting method comprises first performing primary compaction on original point cloud data according to sampling rate, then adjusting categories of remaining data points according to categories of current remaining data points and the number of removed points in k-order neighborhoods of the remaining data points, and carrying out secondary compaction on remaining point cloud according to sampling rate. Compared with a traditional method, the characteristic-maintained point cloud data compacting method has the advantages of being capable of maintaining detailed characteristics of the original point cloud data, avoiding time cost of complex quadric surface fitting and curvature estimation and being capable of effectively avoiding occurrence of holes in compacted point cloud.

The invention provides an improved EMD decomposition-based draft tube pressure pulsation comprehensive evaluation method, and relates to fault feature extraction and state evaluation of pressure pulsation signals of draft tubes of water wheels. According to the method, multi-point pressure pulsation signal features of draft tubes of water wheels are extracted by utilizing an improved empirical mode decomposition (EMD) method, index energy and a multi-scale feature entropy theory, a comprehensive evaluation index is established, and the index is used for evaluating the pressure pulsation degrees of the draft tubes. Though an EMD interval threshold value-based denoising method, background noise interference in the pressure pulsation signals of the draft tubes are removed, intrinsic mode functions IMP expressing different time scales are decomposed through EMD, effective components are extracted by utilizing a correlation coefficient theory, the index energy (IER) is selected to serve as feature parameters so as to carry out feature extraction on the effective components, and a mapping relationship between pressure pulsation energy and a system state confusion degree is established on the basis of the multi-scale feature entropy value theory, so that the pressure pulsation states of the draft tubes are comprehensively evaluated from a new perspective.

The invention discloses an X-band dual-polarization weather radar precipitation estimation method, and relates to a superior integrated wavelet denoising filtering method so as to enhance the phiDP quality and enhance the KDP quality and finally enhance the precipitation estimation accuracy. Three steps of data identification, successive correction and wavelet denoising (called integrated waveletdenoising) are performed on the psiC to acquire the high-quality phiDP, then the KDP is estimated by using the phiDP, and finally precipitation estimation is performed by using the improved KDP through the precipitation estimation equation IDK (KDP, ZDR). The critical point of the whole process is to enhance the phiDP quality from the filtering method so as to enhance the precipitation estimationaccuracy.

The invention relates to a method for extracting characteristics of a three-dimensional model based on a section ray, which aims at the problems of a high calculation complexity, a larger distortion, not having a common research method and being difficult to obtain higher research precision in the existing method for extracting the characteristics of a mode based on contents. The method takes distance sets from specific points on a principal axis section of the model to the model surface as a characteristic description of the model, thus a problem of the model matching is transformed into the problem of comparing a group of distance sets. The method is divided into two sections of characteristic extraction and characteristic matching; when extracting the characteristics, the method can simply and rapidly obtain the model characteristics which have small reconfiguration distortion and can correctly describe a stereochemical structure and structural characteristics of the three-dimensional model. The method has the advantages of simplification, wide applicability, good research quality and high efficiency, therefore, the method has higher practical applicability and wider application range.

The invention discloses a curved surface non-iterative re-gridding method represented by a triangular mesh. The method comprises the steps of when grid redivision is carried out, firstly inputting a grid model to be re-divided, then carrying out the feature point extraction, judging effective characteristic lines, performing model re-sampling by using an LDNI method, performing point cloud quantity optimization and point cloud position optimization, generating a grid by taking a grid radius as a constraint, removing redundant triangular plates by taking a maximum area as a standard, and finally performing grid repair by taking an area, dihedral angles and triangular plate quality as constraints to output. The quality of the triangular plate of the model is quickly optimized in the whole process; a new data structure based on point and triangular plate mixed representation is used, so that the feature information of the resampling point in the original model can be determined during processing. The normal distance is used for replacing the Euclidean distance, so that the detail features can be reserved. The normal information is considered in the repair algorithm, so that the missing part is well restored.

The invention belongs to the technical field of medical devices and discloses an oral clinical simulation teaching system and method. The system is formed by a shock absorber pad, a support, a rotation disc, a multimedia display device, a pitching device, a rotation device, a simulation head model, a simulation oral cavity, simulation teeth, a transmission shaft, a bolt, a mechanic table, a high/low-speed suction-type collection device, a multi-purpose spray gun, an instrument tray, a pin roller and a base. The multimedia display device is arranged on the upper surface of the rotation disc; the rotation disc is arranged on the support; the pitching device and the rotation device are arranged under the simulation head model; the transmission shaft and the bolt are arranged under the pitching device; the bolt is arranged on the mechanic table; the high/low-speed suction-type collection device, the multi-purpose spray gun and the instrument tray are arranged in the mechanic table; and thepin roller is arranged on the base. The system is simple and reliable in structure and low in overall manufacturing cost through simulation design, and meanwhile, allows students to learn oral and dental techniques in a very realistic environment, is high in operability, and can greatly improve students' technical level.

The invention provides an image semantic segmentation method and device and electronic equipment, and relates to the technical field of machine vision, and the method comprises the steps: extracting low-level features and high-level semantic features of a target image through a feature extraction network of a neural network model; constructing the low-level features into a minimum spanning tree structure; inputting the constructed minimum spanning tree structure and high-level semantic features into a tree feature converter in a neural network model to obtain fusion features; and segmenting the target image based on the fusion features to obtain an image segmentation result of the target image. According to the invention, the reliability of image semantic segmentation can be improved.

The invention discloses an electromechanical equipment fault diagnosis method and system. A vibration speed sensor and a vibration acceleration sensor are arranged on different measuring points in thevertical direction of electromechanical equipment bearings; the vibration speed sensor and the vibration acceleration sensor are used for collecting vibration signals; the method comprises the following steps that the electromechanical equipment transmits the vibration signals to a cloud end; the cloud end carries out median filtering and wavelet packet noise reduction treatment on the vibrationsignals, so as to obtain a de-noised signal; the cloud end carries out cepstrum analysis on the de-noised signal, so as to obtain a cepstrum signal; and the cloud performs fault diagnosis according tothe cepstrum signal. The electromechanical equipment fault diagnosis method and system are used for solving the technical problem that the failure diagnosis method cannot detect the imbalance and misalignment of the electromechanical equipment in the prior art, and the accuracy of fault diagnosis is improved.

The invention relates to a sensor performance on-line test device and method based on a multi-threshold wavelet under strong interference. First, a test point environmental parameter signal detected by a sensor is subjected to filtering and collection; then the collected signal is subjected to noise reduction processing through fuzzy multi-threshold wavelet transformation; wavelet decomposition is carried out, wavelet coefficients are obtained, the membership degree of each wavelet coefficient is obtained, the wavelet coefficients with the membership degrees exceeding the preset threshold are rejected, wavelet reconstruction is carried out by utilization of the wavelet coefficients with the membership degrees within the preset threshold, detection data after noise reduction is obtained, sensor characteristic indexes are calculated according to the detection data after noise reduction and environmental parameter values of the test point, and sensor performance on-line automatic test is finished. Noise reduction processing is carried out with combination of filtering and fuzzy multi-threshold wavelet transformation, the signal outline after noise reduction is obvious and clear, no detail signals are lost, fidelity with an original signal is kept, and the signal to noise ratio of the signal is raised obviously.

The invention relates to the technical field of data processing, in particular to a data processing method, a data processing device, a computer readable medium and electronic equipment. The method comprises the steps of obtaining a to-be-processed feature map, and inputting the feature map into an expansion rate calculation layer to obtain an expansion rate table corresponding to the feature map,wherein the expansion rate table comprises expansion rate values corresponding to all pixels in the feature map; and inputting the expansion rate table into an expansion convolution layer to enable the expansion convolution layer to perform expansion convolution processing on the feature map in combination with the expansion rate table so as to obtain feature data corresponding to the feature map. According to the method, the receptive field of each pixel in the feature map can be adaptively and finely adjusted, and feature omission in the feature extraction process is avoided, the sampling density can be improved, and detail features are reserved.

The invention discloses an image denoising method based on an improved orthogonal matching pursuit. The method comprises the following steps that firstly an image is digitalized, and each pixel of theimage is seen as an atom; and then the correlation coefficient is calculated by the residual error r and the absolute value of the inner product among all the atoms of a measurement matrix phi, L values are selected out accordingly and the index values of L values are stored in an index matrix Am, the measurement matrix phi is updated until the condition of the sufficiently low residual error ismet, the finally obtained residual error is less than the last residual error, and the image is reconstructed according to the obtained residual error so as to realize denoising. Sparse coding of signals is performed when Signal sampling is performed, and sparse decomposition and reconstruction of the signals are realized by using the sparsity of the signals so that the detail features of the image can be reserved and the denoising accuracy can be enhanced.

The application relates to the field of computers, in particular to the technical field of vision, and discloses a three-dimensional cartoon face generation method and device, electronic equipment anda storage medium, which can generate a three-dimensional cartoon face which is high in precision and highly similar to a real face based on the real face image. The method comprises the steps: generating a corresponding face-like three-dimensional point cloud model based on a face image; determining a target vertex corresponding to each vertex in the three-dimensional cartoon face model in the face-like three-dimensional point cloud model according to a vertex index relationship, the vertex index relationship comprising a corresponding relationship between the vertex in the face-like three-dimensional point cloud model and the vertex in the three-dimensional cartoon face model; determining coordinates respectively corresponding to each vertex in the three-dimensional cartoon face model based on the target vertex respectively corresponding to each vertex in the three-dimensional cartoon face model; and on the basis of the coordinates of each vertex in the three-dimensional cartoon facemodel, performing fitting to obtain a three-dimensional cartoon face corresponding to the face image.

The invention discloses a stationary Tetrolet transformation algorithm. The stationary Tetrolet transformation is new adaptive Harr class wavelet transformation by four unit squares connected throughsides, and a corresponding filter group is simple and effective. Compared with standard two-dimensional wavelet transformation, the stationary Tetrolet transformation is a new multi-scale geometric transformation tool based on a four-grid joint plate, and anisotropic characteristics in the image can be effectively captured through multidirectional selection. The decomposition and reconstruction algorithm of the stationary Tetrolet transformation is described in detail, and image decomposition by using the stationary Tetrolet transformation is simulated and analyzed. An experimental result shows that compared with the traditional algorithm, the algorithm disclosed in the invention preserves the edge and texture information of the original image and can also effectively acquire good sparse representation, and the defect that Tetrolet transformation algorithm causes blocking effects to image fusion can be eliminated.

The invention provides a target detection method and device, computer equipment, a storage medium and a program product, and relates to the technical fields of computer vision, image processing, artificial intelligence and the like. The method comprises the following steps: segmenting a to-be-detected image to obtain at least two slices, and extracting pixel point features and pixel point context features of each slice through a feature extraction layer to obtain at least two feature maps; detecting the feature map to obtain defect information of the slices; the feature extraction layer comprises a target network layer used for performing pixel point feature extraction on the slices, through the target network layer, down-sampling operation is omitted when the pixel point features of the slices are directly extracted, detail features of an original image are reserved, and even if small targets in a small area range are detected, the small targets can be accurately detected; moreover, a feature extraction layer is also designed to extract pixel point features and pixel point context features, so that the robustness of detection of targets of different sizes is improved, and the accuracy of target detection is further improved.