210 results about "Real time vision" patented technology

The invention discloses a distribution line unmanned aerial vehicle edge end real-time visual identification and target detection method and system based on lightweight AI. The method is based on a YOLOv8 architecture, and constructs a complete lightweight detection framework comprising a feature extractor, an enhancement module and a simplified detection head by introducing a space structure maintaining assembly, a separated large kernel convolution and a weighted reconstruction feature pyramid. A cross-dimension semantic relation model is constructed by utilizing a combined attention structure, multi-level feature fusion is realized by reconstructing a feature pyramid network, model training is performed by adopting a hybrid optimization function and a dynamic sample adjustment mechanism, and the model is deployed on a mobile computing platform after being optimized by a hierarchical knowledge transfer method. According to the method, the detection speed is remarkably increased while high precision is kept, and real-time identification and anomaly analysis of the power line element are effectively realized.

The invention discloses a power transmission channel forest fire risk early warning method and system based on spatial-temporal feature fusion, and relates to the technical field of risk early warning. The method comprises the following steps: constructing a power transmission channel risk grid and fusing historical multi-source data to generate a static flammability background map; the method comprises the following steps: collecting real-time visual and micrometeorological data, and after space-time alignment processing, respectively extracting visual risk features and environmental flammability trend features by using a double-branch space-time feature extraction network; inputting the static background and the dynamic characteristics into a space-time fusion module to obtain a comprehensive forest fire risk index; when the index exceeds a threshold, triggering a forest fire spreading deduction model corrected by combining a power transmission corridor effect, predicting fire spreading, evaluating a line tripping probability, and issuing a graded early warning and emergency strategy; according to the method, the defects of serious data islands and inaccurate early warning are overcome, the crossing from pure fire point monitoring to risk situation dynamic prediction is realized, and the accuracy of forest fire early warning and the intelligent level of power grid prevention and control are remarkably improved.

The invention discloses an edge spraying compensation control method based on intelligent visual feedback. The method comprises the following steps: performing space scanning on a to-be-sprayed workpiece to obtain spraying image data, and establishing a space corresponding relation between a spray gun motion coordinate and a workpiece surface coordinate; registering the real-time spraying image by using the corresponding relation, identifying a geometric boundary line, brightness gradient change and a coating adhesion area of the edge of the workpiece, and generating edge feature data; according to the difference between the edge feature data and the target spraying image, the spraying coverage deviation and the boundary overlapping error are calculated, the coating thickness change trend, the optical density change rate and the boundary direction offset are extracted, and the spraying gun posture adjustment amount, the spraying distance correction amount, the spraying pressure correction amount and the path speed correction amount are generated; the angle, the spraying distance, the pressure and the movement speed of the spray gun are dynamically adjusted according to the parameters; according to the method, real-time visual feedback and self-adaptive compensation control in the spraying process are achieved, and the uniformity and consistency of the coating on the edge of the workpiece are effectively improved.

The invention provides a dynamic data traceability method and system oriented to network security, and relates to the technical field of network security and data governance crossing, and the method comprises the steps: carrying out the real-time visual monitoring of a user interaction interface of a data transaction platform, and capturing the visual behavior information related to data calling; extracting feature parameters based on the visual behavior information to obtain a visual feature data set, and performing structured conversion on the visual feature data set to obtain a structured call log record containing visual behavior associated information; based on the structured call log record, extracting a time sequence, a frequency and a spatial distribution feature of a call behavior as basic indexes; a plurality of feature sampling states are dynamically selected in a monitoring time window, and a dynamic feature evaluation set is constructed based on the feature sampling states. According to the invention, the accuracy and controllability of network security protection in a data transaction scene can be improved.

The invention provides an intelligent driving control method and system for an unmanned ship with a body based on a world model, and relates to the technical field of artificial intelligence, and the method comprises the steps: collecting a real-time visual image, navigation state data and control motion data of the unmanned ship, carrying out the preprocessing and synchronous alignment of the data, and obtaining a real-time visual image; standardized vision, state and action input and a standardized data packet are obtained; the standardized vision, state and action input is mapped into latent representation, autoregression calculation is carried out by predicting a backbone network, and a multi-step latent state and latent action sequence is predicted; and finally obtaining a future state sequence and a future action sequence based on the predicted potential state and potential action sequence. According to the invention, interpretable autonomous intelligent driving is realized.

The embodiment of the invention provides a ceramic diaphragm automatic positioning system based on high-precision visual feedback and closed-loop control, and the system comprises a parameter reference setting module which is used for setting a theoretical reference point and a visual detection parameter; the real-time visual detection module is used for collecting an image of the to-be-stacked film and outputting an image coordinate of the mark point; the core operation module is used for converting the image coordinates of the mark points into actual coordinates based on the visual detection parameters, reading a theoretical reference point of the current to-be-stacked diaphragm and calculating a position error vector between the actual coordinates and the theoretical reference point, and the position error vector comprises a translation compensation amount and a rotation compensation amount; and the motion control and closed loop feedback module is used for receiving the position error vector and generating a control instruction to drive the workbench to move. According to the system, through parameterized reference setting, real-time visual detection, intelligent error calculation and real-time motion compensation, the technical limitation is effectively overcome, and the lamination alignment precision and efficiency are remarkably improved.

The invention provides a self-adaptive contour extraction and target recognition system and method for hand-eye calibration, and the system is characterized in that a hand-eye calibration system which integrates intelligent sampling planning, self-adaptive visual perception, online quality evaluation and closed-loop feedback control is constructed; the core defects of the prior art are overcome. According to the system, through deep fusion of real-time visual perception and motion control in a traditional calibration process, stable, accurate and adaptive extraction and processing of target contour features are realized. The method has the beneficial effects that the calibration robustness and precision in unstructured environments such as illumination variation, shielding and complex textures are remarkably improved, meanwhile, the safety and efficiency of the sampling process are ensured, and the method is suitable for high-dynamic and multi-interference industrial field application.

The invention discloses an unmanned aerial vehicle high-altitude cleaning path precise planning method and system based on visual navigation, and the method comprises the following steps: constructing an initial three-dimensional grid model of a to-be-cleaned building, dividing cleaning regions according to the initial three-dimensional grid model, and setting priorities; the unmanned aerial vehicle collects visual data of a cleaning area through a multi-mode visual sensor, wherein the visual data comprise surface texture, depth information and a stain distribution thermodynamic diagram; performing local correction on the initial three-dimensional grid model based on the real-time visual data to generate a high-precision dynamic three-dimensional grid model; generating a cleaning path by adopting an adaptive grid traversal algorithm according to the priority of the cleaning area and the coverage range of the unmanned aerial vehicle cleaning head; obstacles in the path are detected in real time through a visual feature matching algorithm, and when the obstacles are detected, path re-planning is carried out to avoid the obstacles until all the areas are cleaned. According to the method, the modeling precision and the dynamic adaptability are improved, and the intelligence and the pertinence of path planning are realized.

The invention relates to the technical field of perovskite thin film detection, in particular to a perovskite blade coating thin film defect detection device and method based on real-time visual monitoring, and the device comprises a support, an image acquisition unit, an illumination unit and a self-adaptive adjustment system which are arranged in a glove box and located at a coating machine, a flash evaporation box and a heating stage respectively; the camera and the lens are installed on the adjusting frame, and the first driving piece achieves height adjustment. The light source is installed on the supporting frame and controlled by the second driving piece to ascend and descend. The image analysis unit evaluates the definition, contrast and illumination uniformity of the acquired image, and the control unit automatically adjusts the height and angle of the light source and the position of the camera according to the result so as to keep the image quality stable; according to the scheme, real-time monitoring is carried out in the working procedures of coating, flash evaporation, heat treatment and the like, the comprehensiveness, real-time performance and accuracy of thin film detection are effectively improved, influences caused by environmental fluctuation are avoided, and the accuracy and consistency of defect recognition are guaranteed.

The invention provides a defect detection system and method for bracket processing, and relates to the technical field of defect visual inspection. A visual inspection device is used for collecting a to-be-detected image in real time; performing pixel registration on the to-be-detected image to obtain defect edge uniformity in a defect area on the surface of the bracket, and determining a line form boundary on a defect position on the surface of the bracket according to the defect edge uniformity and an adjacent identification value between adjacent pixel points in defect pixel points; performing texture screening on the defect extension information to obtain gray scale transaction deviations on defect extension abrupt change points in a defect area on the surface of the support, and determining a neighborhood span index in an area to be detected on the surface of the support according to all the gray scale transaction deviations; and determining a pixel incongruous trend according to the grain form boundary and the neighborhood span index, and carrying out visual defect detection on the bracket processing surface. According to the method, accurate real-time visual detection can be carried out on the bracket processing surface defects in a bracket processing dynamic detection scene, so that the accuracy of bracket defect detection is improved.

The invention provides an unmanned aerial vehicle biological control method and system based on real-time visual perception and accurate delivery decision. The method comprises the following steps: collecting farmland environment data through multiple sensors, and performing low-illumination enhancement on a visible light image; utilizing an improved target detection network to synchronously identify diseases and insect pests and position a putting point; the two-dimensional image, thermal infrared and three-dimensional point cloud features are fused to realize accurate positioning of the target; a visual language model is introduced for cross-modal reasoning, target disambiguation and priority analysis are completed in combination with a task instruction, and a collision-free delivery sequence is generated; and finally, planning an optimal flight path, and realizing precise drug delivery through trajectory optimization and closed-loop control. The intelligent level and the operation precision of pest control in a complex farmland environment are improved.

The invention relates to the technical field of intelligent wearing, and discloses an intelligent interaction system and method based on AR glasses. According to the method, an environment scene is intelligently divided, a plurality of interaction areas are generated, then an initial interaction task instruction is received through a user interface, AR glasses are started to execute a task, initial environment data are collected in the execution process, and historical interaction records are obtained. Then, the interaction priority and the attention level of each interaction area are calculated by using the data, so that a customized navigation path is generated. When the AR glasses interact according to the path, position information of the AR glasses is collected in real time, interference source characteristics are obtained, whether the AR glasses enter an interference influence range is detected, if yes, an offset alarm is triggered, whether the AR glasses deviate from a preset path is judged based on the position information, and if not, a path correction command is generated. In addition, real-time visual data can be captured, an abnormal mode or a specific target is analyzed and recognized, and if the abnormal mode or the specific target is recognized, the position information and the corresponding visual data are stored.

The invention relates to the technical field of image updating, in particular to a multi-modal image real-time updating method and system for temporal bone surgery. The method comprises the following steps: acquiring a real-time multi-modal image, performing pixel-by-pixel frequency domain reconstruction optimization, and constructing a frequency spectrum enhanced fusion image; performing reverse geometric transformation compensation on the spectrum enhancement fusion image to obtain a space alignment optimization image; performing real-time visual contrast enhancement on the space alignment optimization image, and constructing a visual enhancement image; carrying out inter-frame difference calculation on the vision enhanced image, carrying out real-time increment updating optimization, and constructing an increment updating image sequence; and performing multi-level cache rendering management and parallel execution based on the incremental updating image sequence. The temporal bone surgery safety and efficiency are improved through real-time and efficient image updating.

The invention belongs to the technical field of medical robots and minimally invasive surgery, and provides a robot control system and method for blue laser vaporization surgery of prostatic hyperplasia. Mapping the nuclear magnetic image volume data to a deformation field under an ultrasonic acquisition coordinate system, and deforming the preoperative nuclear magnetic image to an intra-operative ultrasonic space by using the deformation field to complete image registration; fusing the registered image with a stereoscopic vision system; the spatial depth of the surface of the target tissue is obtained from the endoscopic image so as to supplement navigation information; according to the utility model, the prostate deformation and probe posture change adaptive capacity in an operation is improved, the real-time visual closed-loop regulation and control capacity is improved, and the characteristics of small light spots, shallow heat diffusion, excellent hemostasis and the like of blue laser are combined, so that the vaporization and hemostasis precision in a tiny blood vessel dense area is ensured, and the problems of large tissue trauma and the like are avoided.

The invention discloses a real-time visual language navigation method based on frontier exploration and neural relationship inference. The method comprises the following steps: constructing an initial occupation grid map and a confidence value map; based on the occupied grid map, identifying a boundary between the explored area and the unexplored area, and generating a candidate leading edge waypoint set; a vision-language model is used for detecting instance objects existing in the current visual field, and points closest to the robot are extracted to serve as a candidate instance route point set; calculating visual language comprehensive scores of all waypoints in the candidate frontier waypoint set and the candidate instance waypoint set, and selecting the waypoint with the highest score as a next navigation target; and based on the selected target waypoint, calling a local path planner to generate a collision-free path, driving the robot to move, and updating the occupied grid map and the confidence value map in a rolling manner. Task specific training is not needed, vision and language semantic information can be efficiently fused, and intelligent exploration and target navigation in an unknown environment are achieved.

The invention discloses an automatic rectification method and system for an intelligent patrol point of power transformation equipment, and relates to the technical field of intelligent patrol of the power transformation equipment, and the method comprises the steps: extracting the structural features of a preset patrol point reference map of the power transformation equipment based on a power transformation equipment feature enhancement detection segmentation algorithm, and building a feature library of the patrol point reference map; the method comprises the following steps: extracting real-time visual features of a current frame image in an intelligent patrol process of power transformation equipment; matching the real-time visual features with a feature library of the inspection point location reference map by using a power transformation equipment multi-dimensional feature matching technology, and judging whether the current inspection point location deviates or not according to a matching result; if it is judged that the current patrol point position deviates, based on the visual feature corresponding relation between the reference image and the current frame image, deviation parameters for driving a camera to move are calculated; and generating a camera adjustment instruction according to the obtained offset parameter. The method avoids the hidden trouble of missing faults due to point location offset, and is suitable for large-scale inspection requirements in a complex transformer substation environment.

The invention discloses a feature matching method based on double-branch feature extraction and channel feature enhancement, which is used for improving the feature expression capability and global perception capability of a lightweight feature matching network. The method comprises the following steps: after preprocessing an input image, respectively extracting local details and global structure features through a parallel backbone network and a dynamic channel kernel feature extraction branch; a channel feature enhancement module is used for carrying out progressive strength enhancement on multiple levels; a self-adaptive convolution kernel is generated for each channel through a dynamic channel kernel attention module, and spatial attention modeling of the key area is achieved; a multi-scale feature fusion module is adopted to fuse local and global features, and finally a dense feature map, a key point heat map and a reliability heat map are output; according to the method, the calculation efficiency is ensured, the feature matching precision and robustness in a complex scene are remarkably improved, and the method is suitable for lightweight application scenes of real-time vision SLAM, augmented reality and unmanned aerial vehicle navigation.

According to the photovoltaic power station intelligent cleaning system with cooperation of the unmanned vehicle and the unmanned aerial vehicle, through systematic cooperation of the intelligent control center, the unmanned vehicle and the unmanned aerial vehicle, whole-process intelligence and high efficiency of photovoltaic power station cleaning operation are achieved. A digital twinborn model is constructed through unmanned vehicle environment scanning, a global optimal operation scheme is generated, and a decision foundation of accurate operation is laid; the unmanned aerial vehicle depends on the real-time visual perception and self-adaptive cleaning technology, the crossing from extensive cleaning to precise on-demand operation is achieved, and the cleaning effect and the resource utilization efficiency are remarkably improved; a dynamic docking supply mechanism established between the unmanned vehicle and the unmanned aerial vehicle thoroughly breaks through the cruising bottleneck of the unmanned aerial vehicle, an uninterrupted operation cycle is formed, and continuous self-optimization of an algorithm is realized through a data feedback closed loop. The automation degree, the cleaning quality, the operation efficiency and the economical efficiency of photovoltaic power station cleaning operation are comprehensively improved, and the operation and maintenance problem faced by a large-scale photovoltaic power station is effectively solved.

The invention discloses a pipe fitting self-adaptive positioning and intelligent cutting method and system based on machine vision, and relates to the field of intelligent control, and the method comprises the steps: collecting a pipe fitting image through a multi-view industrial camera, extracting three-dimensional features through a convolutional neural network, and building a complete pipe fitting three-dimensional model; constructing an adaptive positioning model, and solving accurate position coordinates and attitude angles of the pipe fittings by adopting a CNN-LSTM hybrid deep learning algorithm; according to cutting parameter requirements, an optimal cutting path is generated through an A * path planning algorithm, and the cutting robot is controlled to execute operation; images are collected in real time in the cutting process for quality monitoring, and if deviation is found, motion parameters are dynamically adjusted; and archiving and storing the whole-process data, and continuously optimizing the positioning model by applying an incremental learning technology. The method has the advantages that precise three-dimensional modeling and positioning of the pipe fitting are achieved by combining machine vision with deep learning, and intelligent high-quality cutting of the pipe fitting is completed through A * path planning, real-time vision monitoring, PID deviation correction and incremental learning optimization.

The invention relates to the technical field of machining, and discloses a circular knife double-asynchronous collaborative die cutting process and equipment for a multi-glue-layer blue film structure, and the process comprises the following steps: obtaining material parameters and pattern data, and carrying out pretreatment and real-time visual alignment on the multi-glue-layer blue film structure; and the central control module operates an algorithm according to the data, generates and drives the first circular cutter die cutting unit and the second circular cutter die cutting unit to perform asynchronous collaborative die cutting, and peels off waste materials. The system comprises a material conveying module, a visual alignment module, a first circular cutter die cutting unit, a second circular cutter die cutting unit, a central control module and a waste stripping module. By the adoption of the scheme, die cutting alignment precision and efficiency are improved, material loss and production cost are reduced, traditional die cutting errors and material deformation are avoided, cutting quality is optimized, and a technical basis is provided for lean production of electronic product parts.

The invention discloses a digital human live broadcast microphone connection and video generation method based on computer vision, and belongs to the technical field of artificial intelligence, and the method comprises the following steps: 1, real-time vision capture and action expression driving: employing a computer vision tool to carry out face key point detection, facial feature points of lips, eyes and the like of a live anchor or a microphone-connected guest are accurately captured; and then a voice conversion model is matched, microphone-connected audio features and lip key point movement are bound, precise synchronization of the sound and the lip is achieved, and for limb actions, limb joint data can be collected through a monocular camera in combination with a posture estimation algorithm or a simple action capture device and mapped to a digital human skeleton model. According to the digital human live broadcast microphone connection and video generation method based on computer vision, through multi-dimensional technical innovation and systematic design, the problems of insufficient sense of reality, interaction lagging, poor scene adaptability and the like existing in current digital human application are effectively solved.

The automatic gluing robot device is characterized by comprising a six-degree-of-freedom gluing mechanical arm (1), a six-degree-of-freedom gluing mechanical arm (2), a six-degree-of-freedom gluing mechanical arm (3), a six-degree-of-freedom gluing mechanical arm (4) and a six-degree-of-freedom gluing mechanical arm (5), the visual positioning system (3) is used for acquiring three-dimensional contour data of the cover body type part; the control system (4) is in circuit connection with the six-degree-of-freedom gluing mechanical arm (1) and the visual positioning system (3) and is used for coordinating movement of the mechanical arm, gluing parameters and flow control; and the cover body type part positioning device (5) is positioned below the six-degree-of-freedom gluing mechanical arm (1) and is used for feeding and accurately positioning the cover body type parts. By means of the six-degree-of-freedom mechanical arm, real-time visual positioning, intelligent path planning, self-adaptive parameter regulation and control, modular integration and other technologies, high-precision, high-efficiency and high-adaptability automatic gluing operation is achieved.

The invention relates to the technical field of port automation, and discloses a port unmanned crane grabbing pose visual positioning method. The invention aims to solve the technical problems of high system cost, poor environmental adaptability and insufficient positioning robustness caused by excessive dependence of a port unmanned crane on a multi-source expensive sensor, complex calibration deployment process and easiness in signal interference by the environment. According to the method, a double-coordinate reference frame in which a site fixed coordinate system and a crane moving space coordinate system are linked is constructed, and full-process autonomous closed-loop control from path planning to accurate grabbing and unloading is realized by deeply fusing multi-dimensional visual image information. The accurate pose of the crane is calculated through a real-time visual self-calibration mechanism; and a multi-stage visual servo strategy is adopted to guide the grabbing unit to complete high-precision operation. By constructing a set of visual positioning and control system with low cost and high robustness, the autonomy level, the environmental adaptability and the economic efficiency of the system are greatly enhanced.

The invention discloses a processing technology self-correction method and system based on generative AI and real-time visual feedback. The method comprises the following steps: constructing a generative AI process model, receiving a processing demand and generating an initial process scheme; deploying a real-time visual inspection system, and collecting image data in the processing process; extracting processing quality characteristic parameters and comparing the processing quality characteristic parameters with an expected target; dynamically adjusting process parameters based on the characteristic difference; and establishing a mapping relation library of the process parameters and the quality characteristics. The system comprises a process generation module, a visual acquisition module, a feature analysis module, a parameter adjustment module and a knowledge base module. According to the method, the problems of curing and incapability of self-adaptive adjustment of a traditional processing technology are solved, online optimization and self-correction of the processing technology are realized, and the processing quality and efficiency are remarkably improved.

The invention relates to the technical field of data processing, in particular to a logistics loading and unloading monitoring method, device and equipment and a storage medium, and the method comprises the steps: constructing an improved CountVid model, training a track recognition model, and combining multi-angle real-time loading and unloading image automatic processing, thereby achieving the real-time visual monitoring and counting of the whole cargo loading and unloading process, and achieving the real-time monitoring and counting of the whole cargo loading and unloading process. Manual bar code scanning or label identification is not needed, labor cost and errors are reduced, and the automation level is improved; through multi-view dynamic tracking of the cargo trajectory, the full loading and unloading path is covered to avoid a single-view blind area, the problem of counting omission or repetition caused by rapid movement and overlapping overturning of the cargo is solved, and the counting accuracy is improved; by confirming the operation type based on the track data, loading operation and unloading operation can be accurately distinguished, so that the counting result directly corresponds to the type and quantity of the to-be-loaded or to-be-unloaded goods in the transportation task list, errors caused by misjudgment of the operation type in links such as inventory scheduling and order processing are avoided, and it is ensured that statistical data is deeply matched with the logistics business process.

The invention provides a real-time visual feedback man-machine interaction dynamic motion control device, relates to the technical field of robot control, and aims to solve the problems of low capture precision, untimely feedback and poor adaptability in the existing robot motion control technology. The core process is as follows: a demonstrator selects to implement tracking or whole-section capturing dual-mode input actions, a dual-path camera module synchronously captures action information through binocular vision positioning, the action information is analyzed into a three-dimensional action data set through an action trajectory extraction module, a system generates a robot action instruction according to the action information and issues and executes the robot action instruction, and then double-end action data is synchronously captured; the action similarity is calculated through multi-dimensional weighting, standard judgment is achieved in combination with real-time visualization and voice feedback, unqualified actions are automatically corrected, qualified actions are stored and solidified, and by means of closed-loop control, the robot action simulation precision and efficiency are improved, and the action adaptability is enhanced.

The application discloses a kind of path planning methods for realizing Z-axis orientation constraint based on mechanical arm SDK, it is related to industrial robot control field, the path planning methods for realizing Z-axis orientation constraint based on mechanical arm SDK, including target positioning, path planning and analysis, split mechanical arm operating track and dynamically solve Z-axis optimal parameter, visual detection and posture real-time adjustment and transport data record analysis, relies on the real-time visual data of positioning detection system acquisition, realizes Z-axis pose dynamic monitoring, based on solving data, call mechanical arm SDK combined with Cartesian planning algorithm, real-time adjustment Z-axis orientation and end handling component posture and speed, can promptly correct motion error, the Z-axis deviation caused by workpiece deviation, combined with moveit2 frame collision detection capability and SDK fast response capability, realize the path planning method for Z-axis orientation constraint, effectively avoid collision risk, improve the stability and security of mechanical arm operation.

The invention discloses an end-to-end visual reinforcement learning automatic gun insertion charging method for efficient video memory utilization, and belongs to the technical field of robot automatic charging. The method comprises the steps that a system architecture composed of a mechanical arm, a tail end RGB camera and a body sensor is built; constructing a diversified charging scene in an Isaac Gym simulation platform, and generating training data; an end-to-end reinforcement learning model based on a PPO algorithm is adopted, an RGB image and a joint state are used as input, a joint control action is directly output, and direct mapping from sensing to control is achieved; a simulation environment is separated from a reinforcement learning training GPU, video memory utilization is optimized, and the training efficiency is improved; and after training is completed, the model is deployed on a single GPU, and a mechanical arm is controlled to complete a precise gun insertion task according to real-time vision and state information. According to the method, multi-stage processing and complex pose estimation in a traditional method are avoided, and the method has the advantages of system simplification, high adaptability, efficient training and accurate control.

The application discloses a kind of unmanned plane vision enhancement system and method based on binocular rotating polarizer, it is related to computer vision and optical imaging technical field, system includes binocular camera module, rotatable polarizer assembly that can be integrally moved into / out of optical path, mechanism for driving its lifting and rotation and main control unit.The method comprises: after moving polarizer into optical path, in a single acquisition cycle, it is driven to rotate at high speed to multiple preset angles and synchronously acquires binocular image sequence;Process image sequence to calculate Stokes parameters and construct three-dimensional polarization feature space in combination with depth information;The feature space is input into multimodal fusion neural network with original image, and enhanced image, semantic segmentation map and optimized depth map are output.The application realizes the quick and flexible switching of polarization mode and ordinary vision mode, and significantly improves the real-time vision perception and enhancement capability of unmanned plane in underwater, strong reflection and other complex environments through software and hardware depth collaboration.

The invention relates to the technical field of unmanned aerial vehicle control, and discloses an unmanned aerial vehicle accurate landing method based on image recognition, which comprises the following steps: an unmanned aerial vehicle receives a guide signal or a GPS coordinate from a landing platform, and flies to the sky of a landing area; the unmanned aerial vehicle vertically and downwards shoots an image containing a preset landing mark through an airborne camera; preprocessing and feature enhancement are carried out on the shot image, then the image is input to a target detection network, and a center pixel coordinate of a landing mark and contour information of the landing mark in the image are identified; and combining flight height sensor data of the unmanned aerial vehicle according to the corresponding relationship between the pixel size of the identified landing mark and the real-world size of the landing mark. The method is switched to a high-frame-rate image acquisition mode in a short-distance landing stage, realizes sub-centimeter-level position alignment precision through real-time visual servo fine adjustment, effectively eliminates accumulated deviation before final grounding, and provides key space alignment guarantee for stable and accurate landing of the unmanned aerial vehicle.