26 results about "Underwater vision" patented technology

The invention provides a pseudo-supervised diffusion type underwater image enhancement method based on cross-modal semantic guidance, and the method comprises the following steps: obtaining an original degraded image and an actual reference picture of a UIEB, inputting the original degraded image into a multi-modal large model LLaVA to generate a semantic prompt, and selecting a plurality of different existing underwater enhancement methods, the method comprises the following steps: respectively obtaining a plurality of groups of enhanced images, integrating the enhanced images and semantic prompts into image-text pairs, inputting the image-text pairs into a Zip-CLIP model, calculating the similarity of the image-text pairs to generate a plurality of groups of similarity heat maps, inputting the similarity heat maps and high-order semantic information into a special convolutional neural network architecture, generating a pseudo tag as a weak supervision signal for diffusion model training, and carrying out diffusion model training. And a final enhancement result is obtained. According to the method, through the characteristics of cross-modal semantic alignment and gradual reconstruction of the diffusion model, the image detail reduction and structure retention capability is remarkably improved, a more reliable visual basis is provided for an underwater visual task, and theoretical innovation and practical values are both achieved.

The invention belongs to the technical field of underwater vision measurement, and relates to an underwater binocular vision system refraction parameter self-calibration method, which comprises the following steps of: firstly, constructing a space equation containing a water body refractive index, a refraction interface normal vector, an interface distance and a light path coefficient based on a Snell law; constructing a generalized state vector including an environment parameter, a structure parameter and a path parameter; a system residual equation is established by using different-plane light intersection constraint of binocular vision, and nonlinear iterative least square solution is carried out by constructing a generalized Jacobian matrix so as to realize adaptive calibration of an underwater vision system; according to the method, on the premise that water body parameters do not need to be predicted and high-precision mechanical calibration is not needed, the water body refractive index and the system structure parameters are synchronously solved, self-adaptive calibration of the underwater vision system is achieved, and the three-dimensional measurement precision and robustness in the complex water environment are remarkably improved.

The invention belongs to the technical field of computer vision and ocean detection, and discloses an underwater target detection and tracking system, which comprises an underwater image enhancement module used for preprocessing an original underwater video image and outputting an enhanced image; the improved YOLOv11 target detection module comprises an improved YOLOv11, and the improved YOLOv11 is adopted to receive the enhanced image so as to position the underwater target in real time and output a target bounding box; and the target tracking module based on Transform is used for receiving the video clip of the bounding box and extracting the spatial-temporal characteristics, and realizing target association and trajectory prediction based on the target bounding box and the spatial-temporal characteristics. The invention also discloses an underwater target detection and tracking method. According to the system and the method provided by the invention, the accuracy and the robustness of underwater vision tasks including target detection and tracking can be remarkably improved on the premise of ensuring the real-time performance.

The application discloses a method for constructing an underwater static perception model based on terrain fine grid division, and belongs to the underwater visual perception field. The underwater terrain data is finely divided into grids, and the underwater submersible is mapped to the corresponding grid position, so that the static perception model is constructed. In combination with the lake surrounding environment, the lake bottom terrain and the safe navigation depth, a near-non-safe area grid set is constructed, the future grid where the underwater submersible is located is calculated in real time, and if there is an intersection with the non-safe area grid set, an alarm is sent. The model can accurately perceive the underwater static environment and guarantee the safe operation of the underwater submersible.

The invention provides an unmanned monitoring ship based on sonar vision meteorological water quality fusion and a monitoring method thereof, a multi-beam sonar and an underwater camera are used to collect an acoustic image and a visual image of an underwater target, and an embedded main control board is used to realize synchronous registration and fusion of sonar echoes and images. Target detection and recognition are completed based on a deep learning model; meanwhile, a composite water quality probe, an anemorumbometer, a temperature and humidity sensor and an air pressure sensor are used for collecting water body parameters and meteorological parameters in real time to assist in analyzing the water quality change trend; the communication module transmits fusion data to an upper computer in real time through a 5G network, and remote monitoring and intelligent decision making are achieved. According to the invention, through heterogenous information fusion of sonar and underwater vision, the penetrating power of acoustic detection and the high resolution of visual perception are considered, and high-precision detection of fish data in a turbid environment is realized; according to the cross-modal fusion network based on Transform, the target recognition rate and the positioning precision in a complex environment are improved.

This invention discloses an improved YOLO26 underwater target detection method with joint image enhancement. A dedicated underwater image enhancement unit is designed as a pre-task, strengthening image feature representation through a dynamic histogram stretching module and an attention mechanism to output high-quality enhanced images. The YOLO26 network structure is improved by designing a multi-frequency feature reconstruction module and an optimized detection network, achieving collaborative optimization of upstream and downstream tasks between the enhancement and detection units. This method optimizes image quality through enhanced downsampling and upsampling paths. The improved YOLO26 detection network achieves efficient detection and accurate localization of large, medium, and small underwater targets. While ensuring real-time performance and lightweight design, it significantly improves target detection performance under complex underwater conditions, adapting to underwater visual perception scenarios such as underwater search and rescue, marine exploration, and underwater equipment inspection. It provides a practical, generalizable, and easily deployable technical solution for underwater intelligent visual inspection.

The application provides an underwater visual image efficient reconstruction method and system based on implicit entropy learning, comprising: acquiring an input image, the input image being an original image collected by an underwater robot; performing feature extraction on the input image to obtain high-latitude features, extracting auxiliary features, and performing quantization operation on the high-latitude features and the auxiliary features; encoding the auxiliary quantized features to generate an auxiliary feature code stream, decoding and restoring the auxiliary quantized features, decoding and extracting potential context information, and fusing the context information into distribution prediction of the high-latitude quantized features by an implicit entropy modeling module; encoding the high-latitude quantized features to generate a high-latitude feature code stream; decoding and restoring the high-latitude feature code stream to obtain high-latitude quantized features, and performing feature decoding on the high-latitude quantized features by using a decoder to obtain a reconstructed image. The application can solve the problems of insufficient compression performance and poor reconstruction quality of traditional image compression algorithms in underwater complex environments.

This invention discloses an underwater visual-inertial odometry method based on dynamic sparse attention, belonging to the fields of computer vision and robot navigation. An initial visual feature sequence is obtained from two adjacent underwater images. The dark channel map of the underwater image is calculated and encoded as a transmittance-aware embedding. The transmittance-aware embedding is fused with the position encoding to form a 3D position embedding. The 3D position embedding is added to the initial visual feature sequence and input into a visual encoder composed of N stacked dynamic sparse attention modules, outputting a visual feature sequence. An inertial feature sequence is introduced, and a bidirectional cross-attention mechanism is used to achieve deep interaction and alignment between visual and inertial features. An adaptive gating mechanism is used for weighted fusion, and finally, a regression network is used to predict the six-DOF pose. This invention effectively improves the robustness and accuracy of pose estimation in complex underwater environments by introducing physically prior-guided 3D position embedding, dynamic sparse attention, and deep adaptive fusion.

The invention discloses a muddy water image processing method based on image analysis, and relates to the field of image processing. The method comprises the following steps: S1, obtaining an original turbid underwater image shot by an underwater robot; s2, calculating the weight of saturation and significance features in the original image; s3, the saturation weight, the significance weight and the Laplacian weight are normalized, and an enhanced fusion weight map is generated; and S4, inputting the enhanced fusion image into a CNN-GAN filtering model for de-noising processing. According to the method, the problems of low contrast and complex background interference faced by traditional underwater visual detection are effectively solved, and the observation definition, the positioning precision and the measurement reliability of the underwater robot to the underwater target are remarkably improved.

This invention discloses a multi-functional robot for ship maintenance and anti-piracy defense, belonging to the field of marine special robots and intelligent ship operation and maintenance technology. This multi-functional robot comprises two types of isomorphic modular robots: one type is a ship maintenance robot, equipped with underwater visual recognition, hull rust and dirt removal, flexible grinding, and localized spot coating functions, used for routine underwater anti-corrosion and anti-fouling maintenance of ships; the other type is a cutting and welding anti-piracy defense robot, integrating underwater cutting and welding mechanisms, capable of completing underwater damage repairs of ships, while simultaneously using the cutting mechanism to cut and stop pirate boarding boats, external components, and climbing devices, thus possessing both maintenance and anti-piracy defense functions. This invention adopts a robot division of labor and collaborative design. The maintenance end adopts a multi-machine collaborative operation mode, which is highly compatible with the self-developed multi-machine collaborative anti-corrosion and anti-fouling process. The maintenance and defense end realizes the integration of cutting, welding and anti-piracy functions. It has a reasonable structure, strong practicality and easy deployment. It can complete the entire process of underwater operation and maintenance and offshore security operations without stopping the ship or entering the dry dock. It is applicable to underwater maintenance and security defense of military and civilian ships throughout the entire life cycle and has significant market application value and creativity.

The application discloses a cloud-edge collaborative zebra fish three-dimensional trajectory tracking method and system, belongs to the underwater visual perception computing technical field, and aims to solve the technical problem of how to stably and timely track and reconstruct the three-dimensional trajectories of multiple zebra fishes in multiple incubators. The method comprises the following steps: constructing a target recognition model based on a YOLOv5 algorithm, and performing the following on the edge side: obtaining the position information of the zebra fishes through the trained target recognition model, associating the position information and the ID of the zebra fishes through an OC-SORT tracking algorithm, and timely uploading the frame number of the current video image and the ID and position information of each zebra fish in the video image to the cloud side; three-dimensional trajectory construction: calculating the weight of the edges in the directed graph on the basis of realizing trajectory matching through the construction of the directed graph, and realizing the three-dimensional matching of the movement trajectories of the zebra fishes under two different visual angles through multi-view information data fusion.

The application provides a diving mirror for enhancing underwater environment visibility in real time, comprising a camera, a laser lamp, a lens, a semi-transparent display, a fixing band and a mirror frame; the lens is embedded in the mirror frame, the camera is arranged on the upper edge of the mirror frame, the laser lamp is arranged on both sides of the mirror frame, the semi-transparent display is arranged on the inner side of the mirror frame, and the both ends of the fixing band are connected with the both ends of the mirror frame. The diving mirror for enhancing underwater environment visibility in real time is disclosed, the underwater environment is acquired, enhanced and displayed in real time through the underwater high-definition camera and the semi-transparent display, so that the diver can still obtain clear and real images in the case of low illumination, excessive color deviation and serious blur, the problem that the diver's underwater vision is limited is solved, and the diver can more efficiently complete underwater work.

A user's visual capabilities can be evaluated via a virtual reality (VR) system, which can include a VR headset in electronic communication with a computing device. The computing device causes simulated underwater environments, which can include objects, optotypes, and various lighting conditions, to be displayed on the VR headset. Using varying combinations of eye-tracking sensors, eye-tracking cameras, motion-tracking sensors, handheld devices, and microphones, the VR headset collects data about the user as she participates in visual tasks within the simulated underwater environments. Optionally, advanced algorithms in the computing device dynamically alter the simulated underwater environments and the visual tasks and analyze the user's responses to evaluate the user's underwater visual capabilities and her qualification for jobs that require her to work underwater.

The invention discloses a marine ranch shellfish high-precision fishing method and equipment based on multi-source information and visual enabling, and belongs to the technical field of crossing of computer vision, an underwater robot technology and intelligent marine equipment. According to the method, multi-modal input is constructed by fusing optical, sonar and vibration information, and sand surface distortion features are extracted by using deformable convolution and a weighted bidirectional feature pyramid. Further, the relevant local texture of the shellfish is actively enhanced through a deformable local texture enhancement module, and the regional perception robustness is enhanced in combination with graph structure reasoning. And finally, time sequence fusion and probability generation modeling are used, and a high-precision probabilistic positioning result with uncertainty measurement is output. According to the method, the problems of poor underwater visual quality, weak target features and unreliable positioning results are effectively solved, the accuracy, robustness and decision intelligent level of identifying and positioning the buried shellfish by the underwater robot are remarkably improved, and the method is suitable for automatic and accurate catching operation of a marine ranch.

The application discloses an improved YOLO26 underwater target detection method combined with image enhancement, designs a special underwater image enhancement unit as a pre-task, strengthens image feature expression through a dynamic histogram stretching module and an attention mechanism, outputs a high-quality enhanced image, improves a YOLO26 network structure, designs a multi-frequency feature reconstruction module and an optimized detection network, and realizes collaborative optimization of upstream and downstream tasks of the enhancement unit and the detection unit. The method completes image quality optimization through enhanced down-sampling and up-sampling paths, realizes efficient detection and accurate positioning of large, medium and small underwater targets through the improved YOLO26 detection network, significantly improves the target detection performance under complex underwater working conditions while ensuring real-time performance and lightness, is suitable for underwater search and rescue, ocean exploration, underwater equipment inspection and other underwater visual perception scenes, and provides a practical, general and easy-to-deploy technical solution for underwater intelligent visual detection.

The invention discloses an intelligent feeding system and method for penaeus vannamei based on underwater vision and two-factor feedback. The intelligent feeding system comprises an information sensing module, a data processing and decision module and a control execution module, the information sensing module is used for acquiring a prawn image and a residual feed image in a feed table area of the culture pond through an underwater camera; the data processing and decision-making module is used for receiving and analyzing the prawn image and the residual feed image, obtaining the average body length of the prawn according to a prawn body length dynamic measurement model, obtaining the residual feed amount according to a residual feed amount accurate detection model, and sending the residual feed amount to the data processing and decision-making module; inputting the average body length of the prawns and the residual feed amount into a two-factor feeding decision model to generate a feeding decision instruction; and the control execution module is used for receiving and executing the feeding decision instruction and controlling the feeder to complete accurate feeding. According to the method, 'on-demand feeding 'in a real sense is achieved, the feed utilization rate is effectively increased, prawn growth is promoted, and the culture water environment is improved.

The application provides an unmanned monitoring ship based on sonar visual meteorological water quality fusion and a monitoring method thereof, acoustic images and visual images of underwater targets are collected by using a multi-beam sonar and an underwater camera, synchronous registration and fusion of sonar echoes and images are realized by using an embedded main control board, target detection and identification are completed based on a deep learning model; meanwhile, water body parameters and meteorological parameters are collected in real time by using a composite water quality probe, a wind speed and direction instrument, a temperature and humidity sensor and an air pressure sensor, and water quality change trends are analyzed; communication modules transmit fusion data to an upper computer in real time through a 5G network, remote monitoring and intelligent decision-making are realized. Through the fusion of heterogeneous information of sonar and underwater vision, the penetration ability of acoustic detection and the high resolution of visual perception are taken into account, high-precision detection of fish data in turbid environment is realized; the cross-modal fusion network based on the Transformer improves the target recognition rate and positioning accuracy in complex environments.

The invention discloses an image enhancement and feature extraction method and system for an underwater visual odometer, and the method comprises the steps: carrying out the optimization through the comprehensive utilization of the defogging of a physical model, brightness consistency recovery, improved CLAHE, zero reference deep learning enhancement and transformer-based LoFTR feature matching, and fusing the sonar and IMU information; the problems of underwater image blurring, uneven illumination, color distortion, feature scarcity and the like are solved, and the positioning precision and robustness of the underwater visual odometer are remarkably improved; the method does not need a large amount of paired data, is light in model weight, is suitable for real-time deployment on an underwater robot platform, and has a wide application prospect.

The application relates to a background light acquisition method for underwater visual images in an artificial light environment, and belongs to the technical field of underwater robots. The method is characterized in that, starting from the particularity of the working environment of an underwater robot, a depth map is constructed by using information in original underwater images acquired by a charge-coupled camera; then, objects affecting the acquisition of background light in the background area of the original image are segmented out by using the maximum inter-class variance method; finally, the first 0.1% area of the segmented depth map is selected as the farthest area, and on this basis, the average value of the first 0.1% brightest pixels in the area is selected as the value of the background light. Through the application, after the underwater robot acquires images by using the charge-coupled camera, the background light can be quickly solved, the background light data is sent into the main controller of the underwater robot, and a foundation is laid for subsequent underwater image processing.

The invention relates to the technical field of diving equipment manufacturing, in particular to underwater visual aquaculture water quality monitoring equipment which comprises a counterweight frame, the top of the counterweight frame is sleeved with a protective frame, a lifting frame is fixed to the top face of the protective frame, the top of the protective frame is sleeved with a transparent cover, and the outer side of the lifting frame is sleeved with a rotating disc. A high-definition visual monitor is fixed to the surface of the rotary disc, a traction plate is fixed to the surface of the rotary disc, a cleaning frame is fixed to the end of the traction plate, and a sponge pad fixed to the surface of the cleaning frame makes contact with the inner wall of the transparent cover. The system has the advantages that water quality monitoring and underwater visual monitoring functions are integrated, all-directional monitoring of the breeding environment is achieved, more comprehensive and accurate information is provided for farmers, problems in the breeding process can be found in time, corresponding measures can be taken, and breeding benefits are improved.

The invention relates to the field of intelligent aquaculture, discloses an intelligent fish tank monitoring and management system based on edge AI and cloud edge collaboration, and aims to solve the problems that an existing intelligent fish tank lacks the automatic recognition capability of fish states, is easily influenced by underwater visual interference and lacks personalized decision support. The system comprises a sensing layer, an edge computing layer, a transmission layer, a cloud platform layer and an application layer, wherein the edge computing layer executes real-time AI analysis and local immediate response. The cloud adopts two-stage training, and a noise function set containing simulated bubbles and light reflection operators is used in anti-interference pre-training to generate a noise-added image; high-value samples are screened and uploaded through an edge end, and the cloud end executes incremental learning and performs OTA issuing to realize personalization; and providing expert-level diagnosis in combination with real-time multi-modal data and a private knowledge base by utilizing a retrieval enhancement generation technology. According to the method, high-reliability local response, high-robustness identification and intelligent decision support are realized.

The invention discloses an underwater vision-inertia-acoustics SLAM (Simultaneous Localization and Mapping) online calibration and initialization method. The method comprises the following steps: firstly, carrying out time alignment and feature extraction matching on image information acquired by a monocular camera, inertial motion information measured by an IMU (Inertial Measurement Unit) and carrier three-dimensional absolute speed information measured by a DVL (Digital Video Library); based on a motion space consistency principle, optimizing and solving rotation external parameters of the camera and the IMU and gyroscope bias; constructing an acoustic-inertial displacement constraint model, and solving rotation and translation external parameters of the DVL and the IMU; in combination with the DVL speed, IMU pre-integration and calibrated DVL-IMU external parameters, constructing a multi-frame constraint linear equation set to solve a gravity vector and accelerometer bias under a world coordinate system; and finally, estimating a scale factor of the monocular camera and a translation extrinsic parameter between the camera and the IMU based on the calibrated rotation extrinsic parameter and a gravity estimation result. According to the invention, the initialization success rate and the positioning precision of the system in a complex underwater environment can be improved.

The invention discloses an underwater dense object grabbing method based on deep reinforcement learning, and aims to solve the problems of difficulty in perception, low pushing and grabbing cooperation efficiency and insufficient grabbing success rate in an underwater dense scene. Comprising the following steps: acquiring an RGB image and a 2.5 D depth map of a target scene through a camera, and preprocessing; obtaining a target three-dimensional position through coordinate conversion; extracting key features by using a PG-Att-Net network containing an attention mechanism and generating a push / capture Q value; grabbing and pushing action parameters are designed; calculating a global reward through a GPA-Net network, and constructing a multi-dimensional reward function in combination with a local Q value change reward; pushing and grabbing collaborative strategy learning is completed based on an SAC algorithm; and after Gazebo simulation training, transplanting to a physical platform. According to the method, the underwater visual perception precision and the anti-jamming capability are improved, the pushing and grabbing cooperation efficiency is optimized, invalid operations are reduced, the algorithm generalization and the environment adaptability are enhanced, the underwater dense object grabbing success rate is greatly increased, and the method is suitable for actual scenes such as seabed detection.

The invention discloses an underwater vision measurement method and device based on line laser, and the method comprises the steps: determining the pixel distance of a first laser stripe and a second laser stripe on a target detection image according to the slope of the first laser stripe and the slope of the second laser stripe, the target detection image is an optical pattern obtained by shooting a target irradiated by first and second lasers through a camera, the first and second laser stripes are formed by irradiating the target by the first and second lasers respectively, and the first and second lasers are parallel to each other and are parallel to the optical axis of the camera; the distance between the target and the measurement module and the size of the target are determined according to the pixel distance, the measurement module is composed of a first laser, a second laser and a camera, and the first laser and the second laser are used for emitting first laser and second laser which are parallel to each other respectively and are linear lasers. According to the invention, distance measurement and size measurement of a linear target can be realized.