422 results about "Perspective transformation" patented technology

The invention provides an evidence obtaining method and system based on image processing, and the method comprises the following steps: S1, generating a pixel-level depth-of-field distribution diagram of an input image through a multi-scale encoder-decoder network, employing an edge perception optimization layer in a decoding stage, and improving the depth-of-field boundary precision through minimizing a local gradient consistency loss function; s2, performing depth-of-field rationality verification based on an optical imaging physical model, and triggering a first-level tampering alarm by calculating a defocusing fuzzy radius and a gradient direction of a selected region when a difference between the defocusing gradient directions of a target region and a background region exceeds a preset threshold value; and S3, dynamically positioning a key pixel region, identifying a depth-of-field mutation boundary by using an edge detector, calculating by combining local texture complexity, screening a pixel set of which the entropy value is higher than a threshold value and which is located at the mutation boundary, correlating metadata to verify the rationality of the physical size and the spatial position of an object, and eliminating false detection caused by perspective transformation.

Three-dimensional Gaussian splatting mechanisms that initialize a set of 3D Gaussian distributions, un-project pixels from two-dimensional (2D) planes to 3D space by applying queries to the 3D Gaussians at expected un-projected ray depth positions, and splat the 3D Gaussian distributions on the 2D planes based on the expected un-projected ray depth positions.

The invention discloses a multi-view construction personnel tracking method and system based on attention perception. The method comprises the following steps: giving synchronous images from S cameras, and inputting the synchronous images into an encoder for feature extraction to obtain a multi-view feature map; transforming the multi-view feature map into a unified aerial view space by using perspective projection, and aggregating features after projection transformation of all views by using a convolutional layer; inputting the aggregated aerial view features into a decoder for decoding; a cross attention module is introduced, bird's-eye view features of a current frame and an adjacent frame are processed through 3D position coding, instance tokens are extracted as query, keys and values, an affinity matrix is generated through CNN coding similarity, features are propagated through matrix multiplication, and the bird's-eye view features of the current frame are updated. According to the method, the multi-view feature map is projected to the aerial view to realize early fusion, and a cross-frame attention mechanism is introduced, so that the problem of appearance feature distortion caused by perspective transformation is solved.

The invention provides a core photograph automatic correction and cutting method based on YOLOv8-seg and perspective transformation, and belongs to the technical field of image processing, and the method comprises the steps: S1, collecting a core box image; s2, polygonal point sequence labeling; s3, obtaining an optimal segmentation model; and S4, cutting. Aiming at the problems of much manual intervention, low efficiency and poor adaptability in the existing core photograph processing, the invention provides a set of full-automatic end-to-end solution without preprocessing, realizes rapid and stable correction and cutting of the core photograph in a complex scene, relieves the mechanical labor burden of geological personnel, focuses on the core exploration work, and improves the core exploration efficiency. And the practical application requirements of large-scale geological exploration projects are met.

The invention relates to the technical field of scene matching navigation, and provides a multi-mode scene matching navigation positioning method and device for a self-adaptive variable-structure unmanned aerial vehicle. According to the method, geographic coordinates corresponding to all pixel points in an aerial image are determined according to pose information in the aerial image of the unmanned aerial vehicle, and the geographic position of the unmanned aerial vehicle with errors is corrected based on difference information between the pixel coordinates of the central point of the aerial image of the unmanned aerial vehicle and the geographic position corresponding to the center of a camera. Based on the corrected geographic position of the unmanned aerial vehicle, determining a reference image pixel coordinate corresponding to the angular point pixel coordinate of the aerial image, and obtaining a simulated aerial image under the view angle of the unmanned aerial vehicle through perspective transformation so as to determine a homography transformation matrix; according to the method, the satellite reference image pixel coordinates of the interest point in the aerial image are determined based on the homography transformation matrix and the perspective transformation matrix, and finally the longitude and latitude of the interest point are determined, so that the limitation of the traditional method in view angle difference compensation, computing resource constraint and cross-modal processing is solved, and the universality, robustness and positioning accuracy of the system are improved.

The invention discloses an automatic cattle checking method and system based on two-dimensional panoramic vision, and belongs to the technical field of computer vision and intelligent breeding, and the method comprises the following steps: reasonably arranging a plurality of fixed cameras in a cattle farm area, obtaining the internal reference and distortion coefficient of each camera through employing a camera calibration technology, and obtaining the internal reference and distortion coefficient of each camera; performing real-time distortion correction on the collected video frames; on the basis of side-by-side splicing or homography matrix perspective transformation, all paths of corrected images are fused into a complete two-dimensional panoramic image, and a view blind area of a single camera is eliminated; calling a lightweight target detection model on the panorama, and extracting bounding boxes and confidence coefficients of all cattle at one time; sorting according to the confidence from high to low, performing IOU non-maximum suppression de-duplication on the detection frame, and ensuring that the same cattle only counts once in an overlapping region; and the whole process is completed through the edge computing node. According to the invention, efficient, real-time and accurate on-site cattle checking of a large-scale cattle farm can be realized.

The invention provides a suspension bridge main cable wire bulging detection method based on multi-feature fusion, and relates to the field of bridge detection. The method comprises the following steps: detecting the position of a carrier roller by using a deep learning target detection algorithm, constructing a perspective transformation matrix by taking the standard geometric dimension of the carrier roller as a benchmark, converting a squint image into a standard top view, extracting a cable strand area mask so as to construct a width contour vector, carrying out quantitative analysis on drum wire characteristics from six dimensions, and establishing a comprehensive scoring model. Through weighted fusion of multi-dimensional features, intelligent judgment and severity quantitative evaluation of drum wires are realized, the detection efficiency is improved, multi-dimensional feature fusion avoids missing detection of a single feature, good adaptability to a complex field environment is achieved, safety and traceability are improved, manual intervention is reduced through whole-process automatic processing, and the detection efficiency is improved. And the result objectivity is ensured by a multiple judgment mechanism.

The invention provides a method and a device for mapping image coordinates to desktop coordinates, equipment and a storage medium, which are used for improving the positioning precision of a projection touch system in a real environment. The method comprises the steps of performing distortion correction on original gesture video data according to internal parameters of a camera and a lens distortion coefficient to obtain a corrected image frame sequence, and performing interaction space calibration according to the corrected image frame sequence to obtain a distortionless interaction plane, performing perspective change matrix solving on the distortionless interaction plane according to the target vertex coordinate set to obtain orthogonal mapping data, and when gesture key points are detected in the distortionless interaction plane, performing perspective transformation and normalization processing on gesture coordinates of the gesture key points according to the orthogonal mapping data to generate initial desktop control coordinates; and performing display mapping processing on the initial desktop control coordinates according to the physical attribute parameters of the target display to obtain desktop control coordinates.

The invention discloses a computer vision-based railway tool online and offline checking method, which comprises the following steps of: acquiring online and offline and warehouse tool images of a high-speed railway work section, cutting a target area through an area focusing algorithm to eliminate interference, and combining enhancement technologies such as perspective transformation and multi-transformation splicing with a generation method based on SAM and a diffusion model. Constructing a diversity training data set; constructing a text vision multi-modal rotating target detection model, and guiding the model to pay attention to key categories by enhancing feature extraction and fusing text information and visual features of a job schedule; detecting parameters are adjusted by adopting a self-adaptive threshold optimization algorithm, a few-sample detection technology is combined, and after category sensing scale filtering, outlier frame suppression and cross-category non-maximum suppression processing are performed, a result is compared with an operation plan, and a result is output; and based on a sample cutting and screening mechanism driven by false detection and missing detection, high-value samples are mined and supplemented to a training set. The method has the advantage that the safety and efficiency of railway tool management are improved.

The invention discloses a vehicle target detection tracking and trajectory data extraction method based on deep learning, and relates to the technical field of unmanned aerial vehicle aerial photography. The method comprises the following steps: carrying out stable frame processing on an unmanned aerial vehicle video, extracting feature points and feature vectors through an SURF algorithm, matching and screening reliable matching pairs through an FLANN algorithm, calculating a homography matrix through an RANSAC algorithm when a condition is met, carrying out perspective transformation to eliminate jitter, and outputting a stable video sequence; vehicle target detection: introducing an AIFI module to construct an improved YOLOv5OBB model, and outputting vehicle rotation bounding box parameters and categories after training; vehicle tracking and trajectory extraction are carried out, cross-frame tracking is realized based on a DeepSORT model, original trajectory data are preprocessed, and the speed, the acceleration, the additional lane number and the ID of an adjacent vehicle are calculated. According to the method, the problems of video jitter and insufficient detection precision are effectively solved, and the accuracy and continuity of track data extraction in the highway scene are improved.

The invention discloses a writing process analysis method based on computer vision, which relates to the technical field of handwritten character recognition, and comprises the following steps: carrying out perspective transformation on real-time writing data, obtaining a writing image frame sequence, carrying out motion blur correction and pen point coordinate positioning on the writing image frame sequence by adopting an RAFT (Reversible Addition Fragmentation Transform) optical flow algorithm, and forming a writing track data flow; and performing curvature segmentation on the writing trace data stream to obtain discrete stroke segments, and performing spatial topological correlation and geometric structure mapping on the discrete stroke segments to generate a writing stroke topological graph. Through the RAFT optical flow algorithm and the stroke semantic analysis model, the precision of stroke recognition is improved, and efficient and accurate analysis from dynamic visual input to structured character output is achieved.

The invention is suitable for the technical field of automobile auxiliary driving, and provides a quick panoramic image splicing method based on a star flash wireless communication technology, which comprises the following steps of: S1, calibrating internal parameters and distortion coefficients of a camera by adopting a Zhang Zhengyou calibration method, calibrating external parameters by combining perspective transformation, and mapping each view angle image to a unified overlook coordinate system; s2, acquiring an image through a camera supporting a star flash protocol, and transmitting the image to a vehicle body domain controller by using a star flash technology; and S3, after distortion removal and perspective transformation are performed on the received image, a local pyramid fusion strategy is adopted: binary mask marking and multi-band fusion are performed on an overlapping region, original data are reserved in a non-overlapping region, and a panoramic image is finally output. Through combination of high-precision calibration, local pyramid fusion and a star flash communication technology, high-quality and low-delay splicing of vehicle-mounted panoramic images is realized, and the stability and practicability of the system in a complex environment are improved.

The invention provides a panoramic image generation method and device, equipment and a storage medium, and the method comprises the steps: carrying out the single-lens calibration of a wide-angle camera according to a checkerboard calibration method, and obtaining a distortion correction matrix; controlling the wide-angle camera and the array camera to perform time-space synchronous acquisition to obtain an initial panoramic image and a sub-image set; performing nonlinear distortion correction on the initial panoramic image according to the distortion correction matrix to generate a distortionless standardized panoramic image; acquiring sub-image feature information of each sub-image in the sub-image set, and acquiring a projection position, a perspective transformation matrix and a scale transformation matrix corresponding to each sub-image in the standardized panoramic image according to the sub-image feature information; and establishing a panoramic canvas coordinate system according to the standardized panoramic image, projecting each sub-image into the panoramic canvas coordinate system according to the corresponding perspective transformation matrix and the projection position, adjusting the size corresponding to the sub-image in the panoramic canvas coordinate system according to the scale transformation matrix, and generating a panoramic image according to the adjusted panoramic canvas coordinate system.

The invention relates to the technical field of automatic visual detection, and discloses a bottle body label quality detection-based model automatic generation method, which comprises the following steps of: acquiring a multi-angle image of a bottle body, carrying out denoising, enhancement and brightness equalization processing, extracting a curved surface label through target detection and a perspective transformation algorithm, and carrying out image processing on the curved surface label; further generating a distortionless complete label image through an image registration and fusion technology, and taking the distortionless complete label image as a training sample set of a label defect detection model constructed based on a double-branch deep learning network; and performing hot updating on the label defect detection model through a closed-loop feedback and continuous learning mechanism. According to the method, the technical problems of imaging deformation of the curved surface label, diverse defects, difficulty in detection and performance degradation after model deployment are effectively solved, and high-precision, high-robustness and sustainable-evolution automatic label quality detection is realized.

The invention provides a concrete slump detection method based on deep learning, and relates to the technical field of building detection, the concrete slump detection method comprises the following steps: S1, collecting visual data of a concrete slump process, current data of a main shaft of a stirrer and raw material formula data; s2, perspective transformation and binarization processing are carried out on the visual data, and feature extraction and normalization are carried out on the non-visual data; s3, constructing a convolutional neural network and a full-connection neural network, respectively extracting visual and non-visual features, and fusing the visual and non-visual features through an attention mechanism; s4, constructing a slump prediction model based on transfer learning on the basis of the characteristics and historical data; the visual data of the concrete slump process, the current data of the main shaft of the stirrer and the raw material formula data are fused, the deep learning model is used for multi-modal feature extraction and analysis, and compared with a traditional method, the detection error is small, and the detection precision is remarkably improved.

The invention relates to the technical field of water quality detection, and provides a fluorescence detection test paper visualization system and method based on deep learning, and the method comprises the steps: detecting a fluorescence detection test paper image obtained through shooting through a target detection network, and obtaining a test paper main body region, a colorimetric block region and a test paper key angular point coordinate; performing homographic perspective transformation on the main body area of the test paper according to the key angular point coordinates of the test paper, generating a test paper standard view after perspective correction, performing color space standardization correction, generating a mixed deep learning model, and outputting a fluorescence intensity concentration predicted value of the target detection area; generating an interpretable thermodynamic diagram based on the intermediate features of the mixed deep learning model; according to the method, shooting deviation is eliminated through multi-stage image correction, the fluorescence intensity is accurately quantified by fusing CNN and Transform models, the result credibility is improved in combination with a thermodynamic diagram, and the detection automation is comprehensively improved.

The invention discloses a CNN-based high-performance lightweight optimization two-dimensional code recognition method, which belongs to the field of intelligent two-dimensional code recognition and comprises four steps of image preprocessing, lightweight CNN model construction and training, multi-scale feature extraction and decoding post-processing. According to the method, noise is suppressed through improved filtering, two-dimensional code edge details are reserved, and geometric distortion is corrected in combination with perspective transformation; the model parameter quantity and the calculation quantity are reduced through depth separable convolution, a channel attention module is embedded to strengthen key features, and a lightweight CNN model adaptive to the resource-constrained equipment is constructed; complete features are extracted through multi-scale feature fusion, and a decoding result is optimized in cooperation with RS code error correction and repeated recognition voting. Therefore, high-precision and real-time identification of the two-dimensional code in complex environments of uneven illumination, noise interference, distortion and the like is realized on a mobile terminal, an embedded device and the like, and the anti-interference capability and scene adaptability of the method are effectively improved.

The invention discloses a railway station three-dimensional video generation method and system based on a self-adaptive visual angle, and relates to the technical field of passenger station video surveillance, and the method comprises the steps: arranging a plurality of cameras and depth sensors in a railway station, and employing a network clock synchronization mechanism to synchronously collect video streams and point cloud data, generating a real-time sensing data set with a unified timestamp; constructing a static geometric model based on the building information model of the passenger station, and fusing the static geometric model with the real-time data set to generate a dynamic scene point cloud; for a plurality of candidate virtual view angles, counting the coverage degree, the shielding rate, the view angle conversion cost and the key area weight, and calculating view angle scores through a comprehensive scoring function; and selecting the optimal virtual view angle with the highest score, and performing three-dimensional rendering on the dynamic scene point cloud based on the view angle to generate a corresponding three-dimensional video frame. Through a comprehensive scoring mechanism of a virtual view angle, adaptive monitoring coverage is realized, and the monitoring precision and the view angle flexibility are improved.

The invention discloses an instrument data reading and collecting system and method based on image recognition, relates to the field of instrument image recognition, introduces a specially trained Meter-CornerNet deep learning model, and can accurately position four key angular points of an instrument panel from an original image containing perspective distortion and uneven illumination. After the angular points are obtained, an inclined instrument area is corrected into a standard front rectangular view through perspective transformation, and illumination normalization is carried out, so that the interference of the angle and illumination problems on the image quality is improved. And then, digital segmentation and OCR identification are carried out on the processed high-quality image, a result is verified, and finally, an accurate reading is sent to an Internet of Things platform. According to the method, high robustness and high accuracy of the whole system in a complex real scene are ensured through a strategy of first accurate correction and then identification verification.

The invention discloses a license plate character segmentation and comparison identification method, and relates to the technical field of license plate character segmentation and comparison identification, and the method comprises the steps: carrying out the perspective transformation based on the four-corner coordinates of a license plate image, and obtaining a license plate flattening image; reading gray scale percentile positions on the flattened image according to rows, calculating a bright side half width and a dark side half width, determining an unbiased boundary and a character segmentation seam in combination with gradient direction marks, and cutting character image blocks; constructing a single-side index trailing convolution kernel by half widths of two sides, carrying out line-by-line one-dimensional convolution on the standard template to generate a template appearance, selecting candidate characters by adopting normalized cross-correlation, and obtaining a confidence coefficient in combination with a pixel residual error ratio; according to the invention, stable segmentation and high-precision identification can be realized under complex illumination and imaging conditions.

The present invention discloses a method for calibrating a projection light path based on the extraction of chessboard center points. First, the physical world coordinates of the chessboard center point are calculated. The pixel coordinates of the chessboard center point are calculated based on the extracted coordinates of the chessboard corner points. Then, the phase value corresponding to the center point pixel coordinates is mapped to the target surface pixel coordinates of the projector. Finally, the projection light path is calibrated based on the obtained physical coordinates of the chessboard center point and the projector target surface pixel coordinates at the corresponding position. This method first calculates the projector target surface pixel coordinates corresponding to the chessboard center point, and then uses the target surface pixel coordinates and the physical coordinates of the chessboard center point to calibrate the projection light path. Compared with the traditional elliptical calibration plate for calibrating the projector, the complexity of the calculation is reduced, and the result is not affected by the perspective transformation and is more accurate.

The invention provides a pointer instrument reading identification method based on staged detection and inclination over-limit correction. The method comprises the following steps: firstly, acquiring multi-type pointer instrument images, constructing a staged data set and performing data enhancement; training an instrument panel target detection model by using the first part data set, positioning the input image and cutting an instrument panel area; training a key point detection model based on the second part data set, and extracting a starting scale point, a termination scale point, a center point, a pointer tip point and a termination scale value; carrying out normalization processing on the detected key points, judging whether the detected key points exceed an inclination allowable range or not, and executing key point perspective transformation correction on the instrument panel with the inclination exceeding the limit; and finally, combining an angle method and measuring range information to calculate readings. According to the method, only the inclined overrun sample is subjected to geometric correction, full-amount processing is not needed, the calculation power consumption and the processing time can be remarkably reduced while the reading precision is guaranteed in large-scale batch detection, and high precision and high efficiency are achieved.

The invention relates to a pointer instrument reading method and system combining image correction and multi-feature recognition, and the method comprises the steps: obtaining an original image, positioning and cutting a dial region, and obtaining a to-be-processed image; calling the reference dial image, performing feature point identification, and performing perspective transformation on the to-be-corrected image to obtain a standard image; carrying out feature recognition on the standard image, wherein necessary categories comprise a dial plate center point, a half pointer, scale marks and numbers; obtaining pointer tip coordinates based on the half pointer and the dial center point; clustering the identified digits to obtain a plurality of scale values; correlating corresponding scale marks and angle information for each scale value, and performing ascending order arrangement; and screening out two scale marks closest to the tip of the pointer, calculating the relative angle between the tip of the pointer and the two scale marks, and carrying out interpolation to obtain the instrument reading. According to the pointer type instrument reading method and system, the accuracy of the instrument at different inclination angles can be effectively improved.

The invention discloses a real-time video positioning method based on ferromagnetic substance magnetic field signal three-dimensional imaging simulation, and relates to the field of ferromagnetic substance detection in a medical scene nuclear magnetic resonance room, and the method comprises the steps: carrying out the segmentation fitting of a magnetic field gradient through employing a B-spline basis function; performing omni-directional calibration on the array through a rotating platform, and establishing a nonlinear mapping matrix of sensor output and real magnetic field intensity; extracting target shape features; performing optical flow estimation on the video frame, and compensating magnetic field measurement delay caused by target motion; establishing an incidence matrix of the target state vector and the measured value; reconstructing the three-dimensional space distribution of the ferromagnetic substance through an inversion method based on a perspective transformation model; and fusing the three-dimensional imaging result with the real-time video. Magnetic field measurement delay caused by target motion is compensated by performing optical flow estimation on a video frame, and a state estimation convergence speed is improved by estimating a multi-mode parallel state.

The invention discloses an outdoor box video monitoring method, device and system based on multi-dimensional fusion. The system has the advantages that the visible light camera adopted by the main lens collects the physical state of the surface of the box body and the peripheral visible dynamic state, and the short-wave infrared equipment is adopted by the auxiliary lens to collect invisible characteristics and temperature distribution; feature points are extracted to calculate similarity, pixel-level matching is realized through a constructed perspective transformation matrix, and a correlation database is constructed; based on the multispectral data, three types of interaction events of biological interference, natural action and environmental transaction are identified in combination with deep learning, a fault and event association data set is constructed, a weight is assigned, a baseline threshold is set, and early warning is triggered when the threshold is exceeded; after early warning, the starting time and position of an event are positioned, a risk network is constructed to calculate the contribution degree of the event to determine a core risk source, three levels of risks are divided, a traceability map is generated, a differentiated operation and maintenance scheme is generated, monitoring is optimized, and the accuracy and operation and maintenance efficiency of outdoor box monitoring are improved.

The invention provides an industrial material segmentation and size measurement method based on deep learning. The method comprises the following steps: acquiring a to-be-processed industrial material image; inputting the image into the trained segmentation network model, and outputting to obtain a pixel-level segmentation mask; the segmentation network model is a hybrid network based on a U-Net architecture, a pre-trained ResNet50 is adopted by an encoder of the segmentation network model, and a convolutional block attention module (CBAM) is fused in jump connection between the encoder and a decoder; transforming the pixel-level segmentation mask to an aerial view space by using a perspective transformation matrix obtained by pre-calibration to obtain a corrected mask; in the aerial view space, geometric features of the corrected mask are calculated, and the physical size of the material is obtained through conversion according to a preset proportional scale. The device has the beneficial effects that high-precision and automatic segmentation and size measurement of industrial materials can be realized.

The invention discloses a camera calibration method and device, equipment and a storage medium. The method comprises the following steps: acquiring a first image acquired by shooting a calibration plate by a camera; wherein the calibration plate comprises a plurality of angular points; performing inverse perspective transformation on the first image based on the internal reference and the initial external reference of the camera to obtain a second image; detecting angular points in the second image to obtain a first angular point detection result; based on the first corner detection result, the first image and the internal reference, solving to obtain a target external reference of the camera; and calculating an error parameter based on the target external parameter, the internal parameter and the first image, and when the error parameter satisfies a preset condition, determining that camera calibration is completed. According to the technical scheme of the invention, the inverse perspective transformation is carried out on the first image shot and collected by the calibration plate, so that the influence of partial camera distortion can be eliminated, the accuracy of angular point detection is improved, and the success rate and accuracy of camera calibration can be improved.

The invention provides a cowshed drivable area detection method based on fusion of a laser radar and a monocular camera, and the method comprises the steps: synchronously collecting real-time point cloud data and image data, carrying out the matching of the real-time point cloud data and a point cloud map constructed offline, and carrying out the calculation to obtain the pose state of a vehicle in a current map; the method comprises the following steps: inquiring and acquiring key points of a global induction area around a vehicle from a priori map constructed offline, and performing spatial mapping to form an image region of interest; pixel-level classification is carried out through a pre-constructed lightweight semantic segmentation model so as to output and obtain a binary segmentation mask; and mapping the binary segmentation mask from the image pixel coordinate system to the vehicle coordinate system through inverse perspective transformation, and generating a drivable area map under the vehicle coordinate system. According to the invention, through the core thought of priori map guidance, semantic fine recognition and coordinate system unified restoration, the drivable area detection of the cowshed is solved, and a basis is provided for a subsequent path planning module.

The invention relates to the technical field of laser measurement and photoelectric detection, and discloses a laser boresight equipment high-precision calibration method and system based on region segmentation mapping, and the method comprises the steps: firstly collecting a standard grid frosted glass target plate image, extracting grid intersection point pixel coordinates through employing an SURF algorithm, and building a data set corresponding to physical coordinates; dividing the field of view into a plurality of sub-regions, and respectively resolving a perspective transformation matrix mapped from a pixel coordinate system to a physical coordinate system; secondly, collecting a light spot image, performing initial positioning by using a Canny operator and a least square method, fitting a sampling sequence through a Sigmoid function, and extracting edge sub-pixel coordinates in combination with a central difference method; and finally, calling a corresponding transformation matrix according to the sub-region where the light spot is located to obtain physical coordinates, and calculating the light spot jerk value, the maximum diameter and the included angle between the optical axis and the mechanical axis. According to the invention, image distortion is compensated through view field partition mapping, and the calibration precision and detection efficiency of the laser boresight equipment are improved in combination with a sub-pixel positioning technology.

The invention discloses a water surface floating object classification and measurement method based on image recognition, and the method comprises the steps: placing a calibration plate, collecting an image base map and an initial focal length map, and calibrating the pixel feature points of four vertexes of the calibration plate; marking a water body mask and a floating object mask, identifying the water body mask and generating a floating object detection range, and training a floating object identification model by using the floating object mask and a MaskDINO model; the position of the calibration plate is recognized, a current perspective transformation matrix is calculated according to the position of the calibration plate and the physical size of the calibration plate, the current perspective transformation matrix is adopted to map the floating object recognized by the MaskDINO model to a physical plane, and the physical area of the floating object is calculated; identifying the type and position of the floating object based on the zoom image, and matching the initial focal length image with the zoom image identification result; and collecting a floating object pixel detection position of the initial focal length image, calibrating a current perspective transformation matrix to convert a relative physical detection position of the floating object, and performing conversion to obtain the flow velocity of the floating object. The problem of accurate classification of the floating objects is effectively solved.