220 results about "Transformation parameter" patented technology

The invention relates to the technical field of data processing and mode recognition, in particular to a bearing fault diagnosis method based on multi-scale feature fusion, which comprises the following steps: fusing multi-source data such as vibration, acoustic emission and rotating speed, performing angle domain resampling by using rotating speed data, generating a two-dimensional order spectrogram, and stacking to construct a three-dimensional working condition information tensor; a master-slave modulation heterogeneous neural network is adopted, high-dimensional spatial-temporal features are extracted through a main branch three-dimensional convolutional network, time sequence details are extracted from an original sequence through an auxiliary branch one-dimensional convolutional network, affine transformation parameters are generated, and dynamic modulation is achieved on the high-dimensional features; and the output state vector is mapped to a fault evolution knowledge graph, probability prediction is carried out through a graph attention network and by introducing a Monte Carlo discarding mechanism, a probability mean value is calculated as a fault classification result, and the diagnosis confidence is quantified by a probability variance. According to the invention, through multi-scale feature fusion and dynamic modulation, the problem of insufficient feature discrimination caused by scale mismatch under variable working conditions is solved.

The invention discloses a multi-modal image registration method and system based on deformation adaptation and computer equipment, and the method comprises the steps: collecting a plurality of groups of multi-modal images, carrying out the gray standardization, and constructing a diversified registration data set; building a registration network model comprising a pyramid coding module, a deformation adaptive module, a cross-modal interaction module and a registration parameter estimation module; inputting an image pair into the modules in sequence, respectively extracting basic feature mapping, deformation feature mapping and interaction enhancement feature mapping, and finally outputting an estimation conversion parameter matrix; a training process is supervised through a preset loss function, optimal network parameters are selected, and a trained registration model is obtained; in practical application, an image pair to be registered is input into the trained model, a conversion parameter matrix is obtained, and image registration is completed. The multi-modal image registration performance can be effectively improved, and the method still has good robustness and adaptability especially under the condition that serious geometric distortion and significant modal difference exist.

The invention provides an image target intelligent tracking and positioning method and system under Beidou position constraint, and relates to the technical field of Beidou positioning, and the method comprises the steps: obtaining a multi-source heterogeneous data flow, constructing a multi-modal representation model, and building a coordinate system dynamic projection transformation relation; generating a search mask to limit a tracking range by using the Beidou position information; bayesian correction is carried out on a tracking result through double-domain consistency measurement; and dynamically updating projection transformation parameters and feature weights based on residual distribution. According to the invention, deep fusion of vision and Beidou information is realized, and the precision and robustness of target tracking in a complex scene are improved.

The invention discloses a self-adaptive spatial transformation network for cross-medium optical distortion, and solves the problem that the existing spatial transformation network is difficult to consider both global transformation and local nonlinear optical distortion. The network is integrated in a target detection model, and the method comprises the steps that firstly, a parameter prediction network synchronously outputs affine transformation parameters and a water surface fluctuation physical parameter diagram; then, the local distortion grid generation network generates a distortion sampling grid based on the parameter diagram and a predefined water surface fluctuation physical model; meanwhile, the grid generator generates an affine sampling grid according to the affine transformation parameters; then, a grid fusion device fuses the two grids to generate a final sampling grid; and finally, the sampler samples the input image or the feature map according to the final sampling grid to obtain an output image or a feature map. Through a space transformation mechanism guided by physical prior, optical distortion caused by water surface fluctuation can be effectively relieved, and the accuracy and robustness of a target detection model in a cross-medium underwater scene are remarkably improved.

The invention discloses a high-fidelity splicing method and system for an unmanned aerial vehicle image, and the method employs a dual-stage processing strategy, achieves precise positioning through coarse-fine two-stage alignment, calculates an overall transformation parameter through global estimation to complete coarse alignment, refines non-rigid offset through local correction, and achieves high-fidelity splicing of the unmanned aerial vehicle image. The feature weight is dynamically adjusted in combination with self-adaptive enhancement, and geometric smooth constraint is used for limiting regional displacement; smooth weights with continuous boundaries and the weight sum being 1 are generated based on the shape of the overlapping region, images are decomposed to different frequency layers for layer-by-layer fusion, and the system comprises an image acquisition unit, a dual-stage processing unit and an output display unit and can be deployed on the end side of an unmanned aerial vehicle or a ground station. The method is high in alignment precision, can process large parallax and non-rigid deformation, is traceless in fusion, is high in structural fidelity, and meets the real-time processing requirements of airborne or edge equipment.

The embodiment of the invention provides a program conversion method and a compiling device, and relates to the technical field of computers, in the method, an IR file of a GPU program is obtained to serve as a to-be-converted file, for a function parameter of any function in the to-be-converted file, the function parameter is converted into a function parameter matched with a CPU, and the converted parameter is obtained. And determining the IR statement having data dependence with the converted parameter as a statement to be adapted. And for any statement to be adapted, adjusting the statement to be adapted based on the parameter type of the converted parameter and the thread execution mask of the CPU. And compiling the adjusted to-be-converted file to obtain an executable file adaptive to the CPU. The GPU programs written by different high-level languages can be converted into the IR file, so that the conversion is carried out from the dimension of the IR file, and the support for different high-level languages can be realized. Therefore, the limitation is lower.

The invention relates to methods for providing geodetic datum transformation information applicable to a region of the Earth, for use by a navigation satellite system (NSS) receiver and / or a processing entity capable of receiving data therefrom. In one embodiment, the method comprises: generating (s10a) similarity transformation parameters and, for each of a plurality of reference stations in the region, post-fit transformation residuals applicable to the reference station; and sending (s20a) the transformation parameters and output residuals applicable to output points, wherein an output residual is one of the post-fit transformation residuals or derived therefrom; and the distribution of the output points in the region depends on the distribution of the reference stations. The invention also relates to variants of the above-mentioned method, to methods for receiving and processing the geodetic datum transformation information, and to systems, computer programs, and computer program products for carrying out such methods.

The invention provides an automatic trapezoid correction method and system for a projector and a medium. The method comprises the following steps: acquiring attitude information of the projector and spatial information of the projector and a projection surface; under the condition that optical emission parameters of the projector are known, a projection geometric model is established based on the attitude information and the space information, and transformation parameters used for projector trapezoidal correction are generated accordingly. And controlling the projector to execute a trapezoidal correction action based on the transformation parameter. And controlling the projector to project a preset calibration pattern, acquiring an image of the calibration pattern by adopting the image acquisition device after the trapezoidal correction action is executed, extracting image features, and acquiring geometric parameters corresponding to the actual projection picture based on the image features. And determining a calibration error based on the difference between the geometric parameter and the transformation parameter, and generating a compensation parameter according to the calibration error. And writing the compensation parameter into the projector so as to compensate the transformation parameter when the projector performs trapezoidal correction. Through the steps, the projector generates a corresponding correction coefficient.

The invention provides an image automatic exposure method and device, a terminal and a storage medium. The image automatic exposure method comprises the following steps: acquiring a first image, a second image and exposure parameters of the first image and the second image; respectively carrying out a preset number of Gamma transformations on the first image and the second image to obtain a first type of images and a second type of images; determining brightness information entropies of the first type of images and the second type of images with the same transformation parameter and a first transformation parameter corresponding to the maximum brightness information entropy; determining a first update exposure based on the exposure parameter and the first transformation parameter of the first image, and determining a second update exposure based on the exposure parameter and the first transformation parameter of the second image; a joint exposure amount is determined based on the first updated exposure amount, the second updated exposure amount, a first brightness mean value of the first image, and a second brightness mean value of the second image. The method can improve the feature richness of input images of algorithms such as feature matching and depth estimation, and further improves the quality of the algorithms.

The invention provides flexible device alignment detection equipment and a detection method thereof. The flexible device alignment detection equipment comprises a jig platform, a shaping mechanism, a main detection module and a control module. The bearing plane of the jig platform is provided with a positioning reference edge to establish a workpiece coordinate system; the main detection module obtains an unshaped first state reference image and a shaped second state reference image. The control module matches the two-state reference features under a workpiece coordinate system, obtains feature coordinates and solves initial plane rigid body transformation parameters; inversely mapping the second state feature coordinates to the first state to obtain a residual deformation quantity vector, and reconstructing a residual deformation field; and determining and normalizing an adaptive weight coefficient according to the residual modulus value, and solving by adopting weighted least squares to obtain a global alignment deviation of a shaping state relative to a workpiece coordinate system. The equipment can improve the robustness and stability of alignment calculation when in-plane deformation exists.

The invention belongs to the field of mechanical arm control, and relates to a method and device for controlling a mechanical arm by utilizing double-macro lens constraint, and the method comprises the following steps: carrying out system calibration on a double-macro endoscope and the tail end of the mechanical arm to obtain transformation parameters; acquiring an image of a target object in the working space to obtain an original image; performing preprocessing and target detection on the original image to obtain a preprocessed image containing target object features; performing attitude estimation on the preprocessed image to obtain pose information of the target object in a lens coordinate system; a hierarchical iterative approximation strategy of visual feedback is adopted, and a mechanical arm is controlled to drive a double-macro endoscope to gradually approach a target object; the pose information of the target object under the lens coordinate system is converted into a base coordinate system of the mechanical arm; based on the pose information of the target object under the base coordinate system, a mechanical arm joint rotation angle instruction is solved and generated, and a mechanical arm is controlled to complete grabbing operation on the target object; therefore, the fine operation requirement of the manipulator in a narrow working space is met.

The invention provides an animation production method and device, electronic equipment and a storage medium, and relates to the technical field of computers. The method comprises the steps that a skeleton hierarchy is constructed, the skeleton hierarchy comprises a main skeleton and at least one auxiliary skeleton, the main skeleton is a parent skeleton of the auxiliary skeletons, and global transformation parameters of the auxiliary skeletons in a world coordinate system are obtained; the skeleton can be formed by superposing global transformation parameters of the main skeleton in a world coordinate system and local transformation parameters of the auxiliary skeleton relative to the auxiliary skeleton in the world coordinate system; the preset 2D graphic element is bound to the auxiliary skeleton, and the target animation is generated according to the auxiliary skeleton and the preset basic animation, so that differential motion control based on the hierarchical skeleton is realized, the physical motion law of the real world is simulated, the motion freedom degree of the preset 2D graphic element in the target animation can be improved, and the user experience is improved. Diversity of motion trails is achieved, manufacturing efficiency can be improved, and manufacturing complexity can be reduced.

Embodiments of the present application relate to the field of image processing, and particularly relate to an image fusion method, an electronic device, a UAV and a storage medium. The image fusion method comprises: obtaining a first image; obtaining a second image; obtaining a mapping image of the first image based on a first transformation parameter and the first image, the first transformation parameter being used to represent a spatial conversion relationship between the first image and the second image; and weighting and superimposing the second image and the mapping image to obtain a first fusion image. In the embodiments of the present application, the first image and the second image are obtained, and the first image is transformed based on the first transformation parameter, so that the mapping image of the obtained first image is aligned with the second image in terms of spatial structure, that is, the pixel points at the same positions in the mapping image and the second image have a corresponding relationship. In this way, the mapping image and the second image can be fused by directly weighting and adding, and the method is simple.

The application discloses a PDF file editing method and device based on a Web end and related media, and the method comprises the following steps: carrying out hierarchical analysis on document data to obtain pagination resource data, carrying out object analysis on the pagination resource data to obtain a page object model, calculating a unified display matrix of a page space to a device space and rendering to a canvas, superimposing an interactive editing layer to respond to an editing operation and synchronously mapping to the page object model, adapting display transformation parameters of an annotation object, extracting an object operation marked as changed and outputting incremental operation data by page aggregation. The application takes the page object model as a unified data core, runs through the multi-thread cooperative processing of hierarchical analysis, object modeling, unified display matrix transformation and rendering, interactive editing synchronous mapping, annotation display transformation parameter adaptation and incremental operation data output, and makes each link consistent in data structure and coordinate system, so that the precision of editing the PDF file on the Web end is improved.

The invention discloses a radar time error calibration method, related equipment and a medium, and the method comprises the steps: obtaining a first point cloud and a second point cloud, which are obtained through the detection of a same detection target by a first radar and a second radar, in the rotation process of an automatic guided vehicle; performing point cloud registration by using the first point cloud and the second point cloud to obtain a target transformation parameter; determining a target rotation angle of the automated guided vehicle in a process from obtaining the first point cloud to obtaining the second point cloud by using the target transformation parameter; determining a first time difference required by the automatic guided vehicle to rotate the target rotation angle; and determining a time error between the first radar and the second radar by using the first time difference and the second time difference, the second time difference being a difference value between the detection time of the second point cloud and the detection time of the first point cloud. In this way, the accuracy of the calibrated time error of the first radar and the second radar can be improved.

The invention relates to a material image matching method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring edge points of a material template image, normal vectors of the edge points, a material image of a to-be-detected material, and transformation parameters of the material template image and the material image; transforming the coordinates and normal vectors of the edge points based on the transformation parameters to obtain transformed coordinates and normal vectors of the edge points; detecting a connected domain of the material image and a centroid coordinate of the connected domain, and respectively converting the transformed coordinate of the edge point and the coordinate of the pixel point in the material image into a coordinate system taking the centroid coordinate as an original point to obtain a first coordinate of the transformed edge point and a coordinate of the transformed pixel point; and matching the material template image and the material image based on the first coordinate of the edge point, the transformed normal vector and a target edge point corresponding to the edge point in the transformed material image to obtain a matching result. The method is beneficial to improving the image matching efficiency.

The invention is suitable for the technical field of computer application, and provides a one-dimensional code image correction method and device, electronic equipment and a computer readable storage medium, and the method comprises the steps: inputting a to-be-corrected initial one-dimensional code image into an encoder of a preset self-correction neural network, and generating an initial feature map corresponding to the initial one-dimensional code image; inputting the initial feature map into an attitude prediction module of a preset self-correcting neural network, and generating a transformation parameter corresponding to the initial one-dimensional code image; and inputting the initial feature map and the transformation parameters into a decoder of a preset self-correction neural network, and generating a corrected target one-dimensional code image corresponding to the initial one-dimensional code image. Therefore, the initial feature map and the transformation parameters of the one-dimensional code image to be corrected are generated through the preset self-correction neural network, then the one-dimensional code image is corrected according to the initial feature map and the transformation parameters, and the image quality and geometric deformation of the one-dimensional code image are recovered, so that the efficiency and accuracy of one-dimensional code image recovery are improved.

The invention belongs to the technical field of point cloud precise registration, and particularly relates to a point cloud precise registration method for a large-scale building scene. The method comprises the following steps: S1, acquiring a point cloud, and performing hue correction on the point cloud to obtain hue information after hue correction; s2, constructing a hybrid adaptive residual optimization framework according to the point cloud and the hue information, and obtaining an optimal transformation parameter through optimization iteration; and S3, performing point cloud fine registration according to the optimal transformation parameter to obtain a registered point cloud. According to the point cloud precise registration method for the large-scale building scene, high-precision, high-robustness and high-efficiency point cloud precise registration under the conditions of low overlapping rate, strong symmetric structure, illumination variation and geometric noise interference is realized through a hybrid adaptive residual optimization framework fusing geometry and robust hue.

The invention discloses a robust measurable modeling method fusing three-dimensional reconstruction and geographical registration, and belongs to the technical field of three-dimensional modeling. The method comprises the following steps: acquiring source point cloud data generated by three-dimensional reconstruction and target point cloud data with geographic coordinate reference; initial registration is completed based on a Umeyama algorithm; performing iterative optimization on the initial transformation parameters by adopting a Meishan-iterative reweighted least square-Huber weight algorithm, realizing geographic registration through robust standard deviation estimation, Huber weight distribution and convergence judgment, and generating a measurable geographic space three-dimensional model; volume calculation, distance measurement and spatial analysis can be performed based on the model. According to the method, noise and abnormal point interference are suppressed through a robust iterative optimization mechanism, the registration precision is high, the generated model has unified geographic coordinate reference and high-precision measurement functions, the problems that a traditional three-dimensional model lacks geographic reference and is poor in noise immunity are solved, and the method can be widely applied to scenes such as engineering surveying and mapping and space planning.

The present disclosure provides a tower pole component point positioning method and device based on point cloud hierarchical registration, first determining a plurality of resolution levels of target point cloud data and reference point cloud data; then determining the transformation parameters for aligning the target point cloud data to the reference point cloud data; then generating a transformation function based on the transformation parameters, and converting each point in the target point cloud to the corresponding coordinate system of the reference point cloud by using the transformation function; finally, based on the coordinates of the key component points in the reference point cloud, performing neighborhood search in the target point cloud located in the corresponding coordinate system of the reference point cloud, taking the nearest neighbor point of the key component points as the key position points in the target point cloud; and taking the coordinates of the key position points in the target point cloud in the corresponding coordinate system of the target point cloud as the final positioning coordinates of the key position points in the target point cloud.

This application relates to a method, system, and storage medium for multi-scale spatiotemporal fusion of remote sensing data. The method includes: acquiring a first remote sensing dataset with a first spatial resolution, a second remote sensing dataset with a second spatial resolution, and background field data with a second resolution; establishing a spatiotemporal scale transformation model based on the data from both datasets during overlapping periods, generating a parameter set containing scale transformation parameters and quality evaluation parameters; aligning the parameter set with the background field data through coordinate orientation correction and spatial interpolation; dynamically determining continuous fusion weights based on the quality evaluation parameters, and performing weighted fusion of the model output and the background field data to generate preliminary fused data; and finally, performing independent smoothing and denoising processing on the time-series fused data. This method can solve problems such as mismatched parameter spatial matching, spatial discontinuity of fusion results, and poor engineering adaptability caused by high process coupling, achieving spatially continuous, temporally smooth, and easily scaled-up remote sensing data fusion.

This invention discloses a four-way keystone correction method and system for projectors based on a pre-mapped two-dimensional array. The method includes the following steps: Before shipment, a two-dimensional discrete angle grid is constructed, and a pre-stored set of transformation parameters required to correct the trapezoidal distortion of the projected image to a standard rectangle under various angle combinations is collected to construct a pre-mapped two-dimensional array. During projector operation, the corresponding horizontal and vertical attitude angles are collected in real time for filtering, jitter reduction, and discrete step length calibration. Simultaneously, the array index is calculated, and the corresponding pre-stored transformation parameter set is read. If the angles do not perfectly match after discrete step length calibration and quantization, interpolation transformation is performed on the pre-stored transformation parameter sets corresponding to adjacent angle combinations obtained from the index, and the projected video stream corresponding to the projector is corrected to generate the four-way keystone correction result image. This invention enables automatic keystone correction in the vertical and horizontal directions of the projector.

The application relates to the technical field of image detection, and discloses a scale-invariant contour matching method and device based on machine vision and a storage medium, the method comprising the following steps: extracting a target contour from a template and a query image and detecting key points; constructing a multi-scale contour context descriptor for each key point, the descriptor being composed of chord length ratios and chord angle features under multi-scale in a local neighborhood; performing feature matching, estimating scale and rotation transformation parameters according to chord length ratios and chord angles of a matched descriptor pair, and combining key point coordinates to calculate a translation parameter to generate a similarity transformation hypothesis; clustering all hypotheses as points in a high-dimensional parameter space to obtain candidate instances; for each candidate instance, fitting an accurate similarity transformation model and eliminating false matching by using a random sample consensus algorithm through matched point pairs associated with the candidate instance, and outputting final positioning parameters of a target object. The application realizes efficient, accurate and scale- and rotation-invariant contour matching.

The application relates to the technical field of computers and discloses a three-dimensional structure and pose generation method, device, equipment, medium and product. The method comprises the following steps: acquiring a first image and image features of the first image, and acquiring a first three-dimensional structure representation and a first transformation token; the image features, the first three-dimensional structure representation and the first transformation token are injected into a feature interaction module of a third model to obtain a first three-dimensional feature, a first two-dimensional feature and a first transformation parameter; and the first three-dimensional feature and the first two-dimensional feature are aligned by using the first transformation parameter to generate a three-dimensional structure and a camera pose. Thus, joint optimization for three-dimensional structure generation and camera pose estimation is realized, which is beneficial to improving the estimation accuracy of the camera pose and the reconstruction quality of the three-dimensional structure, so that the three-dimensional structure can be aligned with the first image at a pixel level when the three-dimensional structure is projected back to the first image of the original visual angle, thereby ensuring the visual representation consistency of the three-dimensional structure and the first image for a first object.

The application relates to the technical field of deep learning, in particular to a federal learning collaborative modeling method for energy efficiency optimization of a central air conditioning system, which comprises the following steps: acquiring system operation state data, calculating real-time flow data, combining the real-time flow data with supply and return water temperatures, and calculating real-time load values; inputting environmental data into a benchmark generation model to obtain a load mode; inputting system operation data into an energy efficiency model to generate a dynamic remodeling coefficient; through a parameter transformation network, taking the dynamic remodeling coefficient as a transformation parameter of the network to generate a load prediction curve; based on the load prediction curve, solving through an optimal control algorithm to generate a collaborative control instruction, adjusting the operation frequency through the control instruction; using the dynamic remodeling coefficient as a label to train the energy efficiency model and generate updated parameters; after a server aggregates parameters to generate a global optimization model, the global optimization model is distributed, and the energy efficiency model is iteratively updated. Through the federal learning collaborative modeling, the system energy efficiency optimization is realized.

The application belongs to the technical field of computers, and particularly relates to a map element display method, a map element display device, a computer readable medium, an electronic device and a computer program product. The method comprises the following steps: acquiring a three-dimensional model of a map element established in a modeling space, wherein the three-dimensional model of the map element comprises a skeleton line used for representing a main body framework and a model component bound with the skeleton line; acquiring affine transformation parameters used for mapping the skeleton line from the modeling space to a map space; and performing coordinate transformation on the model component according to the affine transformation parameters to obtain a map element used for display in the map space. The application can improve the loading speed and display precision of map content.

The invention discloses a cross-view-angle image retrieval method which is suitable for realizing high-precision matching and positioning of an aerial image of an unmanned aerial vehicle and a satellite tile image under different view angle conditions. The method comprises the following steps: acquiring a satellite image of a target area, performing gridding cutting, and combining an unmanned aerial vehicle aerial image to input a backbone network to extract surface features; unifying the features to 256 dimensions through linear mapping and carrying out batch normalization; calculating query, key and value vectors based on a cross-modal attention mechanism, introducing a global learnable temperature parameter, and generating a gating parameter and an offset item through a multi-layer perceptron to realize feature fusion; utilizing geometric perception fine adjustment to predict six-dimensional affine transformation parameters, and realizing feature alignment through amplitude limit adjustment; and finally, calculating cosine similarity, and retrieving and determining a satellite tile map most similar to the real-time aerial photo, so as to obtain the position of the unmanned aerial vehicle. According to the method, feature alignment difficulty caused by view angle, scale and illumination differences can be effectively relieved, and the accuracy and robustness of cross-view-angle retrieval are improved.

This invention discloses a method, apparatus, equipment, and medium for detecting over- and under-excavation in tunnels, relating to the field of intelligent detection technology in tunnel engineering. The method includes: acquiring raw point cloud data; reconstructing a watertight actual triangular mesh model based on Poisson surface reconstruction; calculating coordinate transformation parameters based on common control points to globally align the actual model with the design model; traversing the vertices of the actual model and calculating the three-dimensional deviation value by determining effective projection points; setting a virtual sliding section along the tunnel axis, analyzing the intersecting contour lines, and integrating to obtain the over- and under-excavation volume; establishing an error propagation model, using Monte Carlo simulation for uncertainty analysis, and generating confidence intervals and confidence levels for the deviation values; finally, generating an over- and under-excavation analysis cloud map and a quantitative analysis report with confidence interval identifiers. This scheme achieves robust deviation calculation, volume integration, and uncertainty quantitative analysis between the design model and the measured model.

Provided are a visual positioning method, a non-transitory computer-readable storage medium and an electronic device. Surface normal vectors of a current image frame is obtained. A first transformation parameter between the current image frame and a reference image frame is determined, by projecting the surface normal vectors to a Manhattan coordinate system. A matching operation between feature points of the current image frame and feature points of the reference image frame is performed, and a second transformation parameter between the current image frame and the reference image frame is determined based on a matching result. A target transformation parameter is obtained, based on the first transformation parameter and the second transformation parameter. A visual localization result corresponding to the current image frame is output, based on the target transformation parameter.

According to an embodiment, an information processing apparatus includes one or more hardware processors configured to function as a memory control unit, a transformation unit, a first convolutional neural network (CNN), and a second CNN unit. The memory control unit reads a first stride parameter used for controlling an output resolution and a first dilation parameter used for controlling an input resolution from a memory device. The transformation unit transforms the first stride parameter to a second stride parameter and transforms the first dilation parameter to a second dilation parameter by using a transformation parameter. The first CNN unit executes first CNN processing of a feature vector by using at least the second stride parameter. The second CNN unit executes second CNN processing with an output vector of the first CNN unit as an input by using at least the second dilation parameter.