Water surface garbage automatic identification and cleaning device and method

Abstract

The invention discloses a water surface garbage automatic identification and cleaning device and method, and relates to the technical field of environmental protection equipment, the water surface garbage automatic identification and cleaning device comprises a hull platform, a sensing system, a calculation and control system, a garbage cleaning execution system and an energy and return voyage system; an optical sensing module in the sensing system is used for acquiring water surface target data, and a positioning module is used for positioning a ship body; the calculation and control system is provided with a water surface garbage recognition model and used for processing data collected by the optical sensing module in real time and outputting garbage recognition information. The garbage cleaning execution system is controlled by the calculation and control system and is used for executing garbage salvage and collection operation; the energy and homeward voyage system is used for monitoring the electric quantity of the device and the capacity of the garbage bin and conducting homeward voyage judgment and execution according to the electric quantity of the device and the capacity of the garbage bin, and the problem that in the prior art, automatic and intelligent operation of the whole process from garbage recognition and salvage to endurance management cannot be achieved during water surface garbage cleaning can be solved.

Description

Technical Field

[0001] This invention relates to the field of environmental protection equipment technology, and in particular to an automatic identification and cleaning device and method for surface debris. Background Technology

[0002] With the advancement of urbanization and the increase in human activities, the problem of floating garbage pollution on the surface of rivers, lakes, and reservoirs has become increasingly serious. This kind of garbage not only damages the aquatic landscape and ecological environment, but may also threaten navigation safety, the ecological balance of aquatic bodies, and coastal facilities.

[0003] Currently, the cleanup of surface debris mainly relies on two methods: manual retrieval and semi-automated cleaning vessels. Manual retrieval involves workers operating boats to conduct visual searches and manually collect debris. This method has significant drawbacks, including high labor intensity, low efficiency, limited coverage, high costs, and high safety risks for personnel operating in complex aquatic environments, making it difficult to meet the needs of large-scale, continuous debris cleanup. Semi-automated cleaning vessels are typically equipped with simple collection devices such as conveyor belts and robotic arms, but their identification and decision-making capabilities heavily depend on manual remote control or pre-set fixed routes. They lack intelligent perception of the aquatic environment and cannot autonomously and in real-time identify debris and dynamically plan the optimal cleaning path. Therefore, their operational efficiency, level of intelligence, and adaptability to complex environments remain insufficient.

[0004] In recent years, with the development of machine vision and artificial intelligence technologies, some solutions have emerged that attempt to use cameras for water surface debris identification. However, these technologies based on ordinary RGB visible light imaging have fundamental limitations when dealing with complex and ever-changing water surface environments: for example, their discrimination ability is weak, making it difficult to effectively distinguish objects with similar colors and textures but different materials, such as green plastic bottles and duckweed, or dark plastic bags and shadows or silt; their environmental robustness is poor, and their recognition effect is easily affected by water surface reflections, ripples, weather changes, and lighting conditions, resulting in unstable recognition rates and high false positive and false negative rates.

[0005] The aforementioned problems make it difficult for existing technologies to achieve high-precision, robust, and real-time identification of debris in complex aquatic environments, thus failing to support truly fully automated and intelligent cleaning operations. Specifically, there is an urgent need for a systematic solution that can reliably identify debris in complex environments, intelligently and dynamically plan paths, effectively avoid obstacles, and achieve autonomous endurance management. Summary of the Invention

[0006] This invention provides an automatic identification and cleaning device and method for water surface debris, which can solve the problem that existing technologies cannot achieve fully automated and intelligent operation of water surface debris cleaning from debris identification and retrieval to continued operation management.

[0007] To address the above problems, the present invention provides an automatic water surface debris identification and cleaning device, comprising: The hull platform is equipped with a power steering system and a waste collection compartment. A perception system is installed on the hull platform. The perception system includes an optical sensing module, a positioning module, and an obstacle avoidance module. The optical sensing module is used to acquire water surface target data, and the positioning module is used to locate the hull. The computing and control system is equipped with a surface debris recognition model, which processes the data collected by the optical sensing module in real time and outputs debris recognition information. A waste collection and disposal system is installed on the ship's platform. The waste collection and disposal system is controlled by the computing and control system and is used to perform waste retrieval and collection operations. The energy and return-to-home system, including the battery pack and battery management system, is used to monitor the device's power level and waste bin capacity, and to determine and execute the return-to-home decision based on the device's power level and waste bin capacity.

[0008] The present invention provides an automatic identification and cleaning device for surface debris, which, compared with the prior art, has the following beneficial effects, but is not limited to: The device first utilizes a sensing system integrating optics, positioning, and obstacle avoidance to comprehensively perceive its environment and its own status, providing a data foundation for intelligent operations. Secondly, a computing and control system equipped with a recognition model processes the data in real time and outputs waste identification information, enabling intelligent decision-making. This system then controls the cleaning actuator to accurately complete the retrieval and collection, forming a closed loop of perception-decision-execution. Finally, an energy and return system monitors battery power and storage capacity and autonomously manages the return process, ensuring operational continuity and safety. The entire solution achieves full automation from environmental perception, intelligent identification, autonomous cleaning to endurance management, significantly reducing manual intervention and improving the efficiency and intelligence level of water surface waste cleaning.

[0009] Preferably, the optical sensing module includes a spectral imaging sensor, which is used to acquire reflectance data of water surface targets in multiple continuous or discrete bands to collect raw hyperspectral image data.

[0010] Preferably, the execution of the computing and control system includes the following steps: The original hyperspectral image data is preprocessed; Feature band extraction is performed on the preprocessed image; The image after feature extraction is input into the water surface debris recognition model to obtain pixel-level segmentation maps or target-level detection boxes.

[0011] Preferably, the feature band extraction employs at least one of principal component analysis, independent component analysis, or linear discriminant analysis.

[0012] Preferably, the water surface debris recognition model is trained through the following steps: Obtain hyperspectral image samples from labeled and unlabeled datasets, and generate corresponding labels for the hyperspectral image samples from the labeled datasets; Perform one data augmentation on the labeled dataset with labels, and perform K data augmentations on the unlabeled labeled dataset; For the unlabeled data after K augmentations, the model predicts and calculates the average probability distribution to generate pseudo-labels, and then sharpens the pseudo-labels. The augmented labeled data is mixed with the sharpened unlabeled data to generate mixed data; Calculate the cross-entropy loss of labeled data and the mean squared error loss of unlabeled data, and then sum the two by weight to obtain the overall loss function; The model parameters are optimized based on the overall loss function until the model converges.

[0013] Preferably, the computing and control system is further configured to: Based on the identification information of multiple consecutive frames and the positioning data provided by the positioning module, a heat map of water surface garbage distribution is generated and dynamically updated. Based on the heat map of surface debris distribution and the environmental information provided by the obstacle avoidance module, the cleaning route is planned and adjusted in real time.

[0014] Preferably, the heat map of surface waste distribution is mapped using a geographic coordinate system and uses different colors or numerical gradients to characterize the distribution density, type, or cleanup priority of the waste.

[0015] Preferably, the obstacle avoidance module includes a millimeter-wave radar.

[0016] Preferably, the energy and return-to-home system is configured as follows: Real-time monitoring of the remaining power of the battery pack and the loaded capacity of the waste collection bin; When the remaining power is lower than the first preset threshold, or the loaded capacity reaches the second preset threshold, it is determined that a return trip is required. The computing and control system is triggered to plan the optimal path back to the preset base, and the power steering system is controlled to drive the ship platform to return autonomously along the path.

[0017] Preferably, this application embodiment also provides a method for automatic identification and cleaning of surface debris, applied to the device described in any of the above claims, the method comprising: Data on water surface targets are collected using an optical sensing module; The collected data is processed in real time by a surface debris identification model in the calculation and control system, and the debris identification information is output. Based on the identification information and positioning data from multiple consecutive frames, a heat map of surface debris distribution is generated and updated. By integrating environmental perception information from heat maps and obstacle avoidance modules, a cleaning route is planned. Control the hull platform to navigate along the planned route and control the garbage collection system to perform salvage operations in areas with dense garbage; The system monitors the operational status through an energy and return-to-home system, and automatically controls the device to return to home when the return-to-home conditions are met. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the automatic identification and cleaning method for surface debris in water according to an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the execution steps of the computing and control system according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the training steps of the water surface debris recognition model according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the configuration structure of the energy and return system according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings showing multiple embodiments according to this application. It should be understood that the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments described in this application without creative effort will fall within the scope of protection of this application.

[0021] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used in the description of this application is for the purpose of describing specific embodiments only and is not intended to limit this application; the terms "comprising," "including," "having," "containing," etc., in the description, claims, and accompanying drawings of this application are open-ended terms. Therefore, "comprising," "including," or "having" refers to, for example, a method or apparatus having one or more steps or elements, but is not limited to having only these one or more elements. The terms "first," "second," etc., in the description, claims, or accompanying drawings of this application are used to distinguish different objects, not to describe a specific order or hierarchy. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0022] In the description of this invention, it should be understood that the terms "upper," "lower," "left," "right," "front," "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0023] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "attachment" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0024] It should be emphasized that when the term "comprising / including" is used in this specification, it is used to explicitly indicate the presence of the stated feature, integer, step, or component, but does not exclude the presence or addition of one or more other features, integers, steps, parts, or groups of features, integers, steps, or parts.

[0025] like Figures 1 to 4As shown in the figure, an automatic water surface debris identification and cleaning device provided by an embodiment of the present invention includes a hull platform, a sensing system, a computing and control system, a debris cleaning execution system, and an energy and return system. The hull platform is equipped with a power steering system and a debris collection bin. The sensing system is installed on the hull platform and includes an optical sensing module, a positioning module, and an obstacle avoidance module. The optical sensing module is used to acquire water surface target data, and the positioning module is used to locate the hull. The computing and control system is equipped with a water surface debris identification model, which is used to process the data collected by the optical sensing module in real time and output debris identification information. The debris cleaning execution system is installed on the hull platform and is controlled by the computing and control system to perform debris retrieval and collection operations. The energy and return system includes a battery pack and a battery management system, which is used to monitor the device's power level and the debris bin capacity, and to determine and execute the return operation based on the device's power level and the debris bin capacity.

[0026] Within the aforementioned overall structure, the hull platform integrates a power steering system consisting of an electric motor-driven propeller and a rudder mechanism, as well as a waste storage compartment. Four core modules—sensing, decision-making, execution, and energy—are systematically arranged on the hull platform.

[0027] Specifically, the perception system is an integrated sensing unit comprising at least three functional modules: an optical sensing module, a positioning module, and an obstacle avoidance module. The optical sensing module is positioned facing the water surface and its function is to acquire data on water targets, i.e., to collect raw information about the water surface state. The positioning module continuously acquires the device's own position information to achieve vessel positioning. The obstacle avoidance module detects obstacles ahead of the vessel.

[0028] The computing and control system uses an embedded computer or microcontroller as its core hardware, in which a pre-trained model for identifying surface debris is deployed. This computing and control system communicates with the optical sensing module of the perception system to process the data collected by the optical sensing module in real time. By running the identification model, the system can analyze and calculate the input data and ultimately output debris identification information, which includes at least the presence and location of the debris.

[0029] The garbage collection system performs physical retrieval, picking up and collecting garbage from the water surface. Its structure can be set up using existing devices. The energy and return system provides power and manages endurance, ensuring the stability of energy supply and task completion during long-term operation. The battery management system not only monitors key parameters such as battery voltage, current, and temperature in real time, but also has overcharge, over-discharge, overcurrent, and short-circuit protection functions, effectively extending the battery pack's lifespan. The waste collection bin is equipped with a capacity sensor that can accurately detect the volume or weight of the currently loaded waste and use calculations and the control system to determine whether to perform a return.

[0030] In this embodiment of the application, the optical sensing module includes a spectral imaging sensor, which is used to acquire reflectance data of water surface targets in multiple continuous or discrete bands to collect raw hyperspectral image data.

[0031] In the above structure, the spectral imaging sensor is installed on the bow or a liftable mast and continuously scans the water surface in front at a certain pitch angle. The hyperspectral image data it collects contains spectral information of dozens to hundreds of continuous bands from visible light to near infrared. By analyzing the absorption and reflection characteristics of the target object at a specific wavelength, it can effectively distinguish similar objects that are difficult to distinguish with the naked eye. For example, it can accurately distinguish between green plastic bottles and natural duckweed, or between dark plastic bags and shadows and silt on the water surface.

[0032] To ensure data quality, the optical sensing module can also integrate auxiliary light sources and polarizers to reduce the impact of strong water surface reflection, wave interference, and lighting conditions under different weather conditions on imaging, thereby further improving the stability and reliability of target data acquisition in complex environments.

[0033] In this embodiment of the application, the execution of the computing and control system includes the following steps: The original hyperspectral image data is preprocessed; Feature band extraction is performed on the preprocessed image; The image after feature extraction is input into the water surface debris recognition model to obtain pixel-level segmentation maps or target-level detection boxes.

[0034] The above steps include preprocessing the original hyperspectral image data by smoothing, standardizing, and normalizing it to remove noise, correct spectral drift and interference caused by illumination inhomogeneity, and ensure data consistency and reliability.

[0035] Feature band extraction is based on the spectral fingerprint characteristics of different types of waste. Principal component analysis (PCA), continuous projection algorithm (SPA), or band selection algorithm are used to screen out the key band combinations that contribute the most to waste identification from high-dimensional spectral data. This reduces the data dimensionality and computational load while retaining effective information, thus speeding up subsequent model inference.

[0036] The surface litter identification model employs a deep learning network architecture trained using a MixMatch semi-supervised learning method, such as improved U-Net or DeepLab semantic segmentation models, or object detection models like YOLO and Faster R-CNN. During training, the model not only utilizes a small number of labeled hyperspectral litter samples for supervised learning but also enhances its generalization ability and recognition accuracy across different types of litter, lighting conditions, and water quality backgrounds through data augmentation and pseudo-label generation on a large number of unlabeled samples. The model's pixel-level segmentation map accurately delineates the litter's contour boundaries, while the target-level detection box provides the litter's center coordinates, bounding rectangle size, and class probability, offering precise target information for subsequent heatmap generation and route planning.

[0037] In this embodiment of the application, the feature band extraction employs at least one of principal component analysis, independent component analysis, or linear discriminant analysis.

[0038] Principal component analysis (PCA) maps high-dimensional spectral data to a low-dimensional space through orthogonal transformation, retaining the principal components with the largest variance in the data. This removes redundant information while preserving the key features of the original data to the greatest extent possible. It is particularly suitable for reducing the dimensionality of spectral data and highlighting the differences between the principal components.

[0039] Independent component analysis (ICA) aims to separate statistically independent source signals from mixed signals. It can effectively extract spectral features with non-Gaussian distribution characteristics and has advantages in identifying garbage types with unique spectral fingerprints.

[0040] Linear Discriminant Analysis (LDA) focuses on finding projection directions that maximize the discriminative power between different categories. By optimizing the ratio of intra-class divergence to inter-class divergence, it enhances the model's ability to distinguish between different types of waste. In practical applications, depending on the specific waste type, water quality conditions, and required recognition accuracy, a single method or a combination of methods can be selected for feature band extraction to achieve optimal preprocessing results and provide high-quality input data for subsequent waste identification models.

[0041] By evaluating the three methods described above, the method with the highest accuracy was selected for feature wavelength extraction. For example, in scenarios where plastic waste is the primary target for identification, a combination of continuous projection algorithm and independent component analysis (ICA) can be prioritized. Continuous projection algorithm reduces redundancy between bands, while ICA separates the spectral signals unique to plastic. When multiple mixed wastes need to be identified simultaneously, a combination of principal component analysis (PCA) and linear discriminant analysis (LCA) can more efficiently reduce dimensionality and enhance category differences. After feature band extraction, the data dimensionality can be compressed from hundreds of bands to 10-20 key bands. While ensuring that the loss in identification accuracy is less than 5%, the model inference speed is increased by 3-5 times, meeting the computing power requirements of real-time operation of the device.

[0042] In this embodiment of the application, the water surface debris recognition model is trained through the following steps: Obtain hyperspectral image samples from labeled and unlabeled datasets, and generate corresponding labels for the hyperspectral image samples from the labeled datasets; Perform one data augmentation on the labeled dataset with labels, and perform K data augmentations on the unlabeled labeled dataset; For the unlabeled data after K augmentations, the model predicts and calculates the average probability distribution to generate pseudo-labels, and then sharpens the pseudo-labels. The augmented labeled data is mixed with the sharpened unlabeled data to generate mixed data; Calculate the cross-entropy loss of labeled data and the mean squared error loss of unlabeled data, and then sum the two by weight to obtain the overall loss function; The model parameters are optimized based on the overall loss function until the model converges.

[0043] In the above identification steps, a water surface debris identification model can be constructed using the Mix Match semi-supervised learning method. The main processes of this identification model include: The collected dataset was divided into two parts: a smaller labeled dataset and a larger unlabeled dataset. For the labeled dataset, professionals performed pixel-level or target-level annotations on the garbage targets in the hyperspectral images, defining categories (such as plastic bottles, plastic bags, leaves, foam blocks, etc.) and generating labels.

[0044] Data augmentation: For labeled data, perform an augmentation operation. =Au( For unlabeled data, perform K augmentations. =Au( This step increases the training dataset, making it as diverse as possible, thus enhancing the generalization ability of the trained model. For unlabeled data, K augmentations are performed. This step increases the training dataset, making it as diverse as possible, thus enhancing the generalization ability of the trained model.

[0045] In the above: This refers to labeled batch hyperspectral image data, with the subscript 'b' representing the batch. Au() is a data augmentation operation function, such as image rotation, cropping, brightness adjustment, etc. This is unlabeled batch hyperspectral image data; =Au( This involves performing one data augmentation on a single batch of labeled data to obtain the augmented data. =Au( To perform K data augmentations on a single batch of unlabeled data, where K is a manually set hyperparameter.

[0046] Label guessing: Based on the identification results of the unlabeled data model after K enhancements in the previous step, a "pseudo" label is guessed, and the average is used to predict the label for the unlabeled data.

[0047] In the above: The probability distribution of the generated pseudo-labels; K represents the number of augmentations for the unlabeled data; Pm() is the model prediction function; y represents the waste category, which can include items such as plastic bottles and plastic bags; θ represents the parameters of the model.

[0048] The "pseudo" labels obtained after label guessing are sharpened to reduce the entropy of the "pseudo" labels guessed in the previous step, highlight the categories with high probability, and make the predicted probability distribution more obvious.

[0049]

[0050] In the above: The probability distribution of the input is denoted by the pseudo-label q in this formula; T is a temperature hyperparameter. The smaller the T value, the steeper the probability distribution and the more certain the pseudo-label. L represents the total number of waste categories; This represents the target category whose probability is currently calculated after sharpening; It is the traversal index for the summation operation.

[0051] Use the first The original probability of class 1 raised to the power of T, divided by the sum of the original probabilities of all classes raised to the power of T, yields the result of the first class. The probability after class sharpening is calculated in a way that highlights high-probability classes while reducing the proportion of low-probability classes.

[0052] The mixture will first be formed by identifying and predicting labeled data. Sum of probabilities Put in =(( , b∈(1,...B)).

[0053] Unlabeled data and the probability after sharpening Put in =(( , ); b∈(1,...B)k∈(1,...K)).

[0054] Will and The data are mixed together and then randomly rearranged using a shuffle process to obtain dataset W.

[0055] and W via MixUp ( W) Output tag data , and W via MixUp ( W) Output tag data .

[0056] In the above: Given a labeled data set, the elements are ( , ), yes The true label; This is an unlabeled data set, with elements being ( , ), These are sharpened pseudo-tags; B represents the total number of batches with tagged data; The fusion coefficient is obtained using a probability distribution:

[0057]

[0058]

[0059]

[0060] In the above: and This represents the input data vector. and It's similar to one-hot encoding, where categorical variables are represented as binary vectors, addressing the issue of classifiers struggling with discrete data. The label is λ, a positive number between [0,1], representing the mixing ratio between two samples. The weighting factors... The hyperparameter α is obtained through Beta sampling.

[0061] In this case, λ follows a Beta distribution. These are hyperparameters used to control the mixing ratio; Used to adjust the mixing ratio to ensure that the weights of the mixed data are more balanced; To mix two data samples and New samples were obtained ; The labels of the two samples are combined to obtain a new label. .

[0062] Calculate the loss, and calculate different loss terms.

[0063]

[0064]

[0065] H It is the classification cross-entropy, T, K, α, and λ are hyperparameters, which are parameters that are manually adjusted before or during training. The final overall loss function is a weighted sum of the two. This is the weighting factor of the unsupervised learning loss function, which can be adjusted as a hyperparameter.

[0066]

[0067] This represents the number of samples in the mixed labeled dataset. The loss value is the model error calculated based on hyperspectral images labeled with waste categories, such as the error of a sample labeled as a plastic bottle being misclassified as a leaf by the model. The loss value is for unlabeled data, and the model error is calculated based on unlabeled but pseudo-labeled hyperspectral images, such as the error of a sample with a pseudo-label of plastic bag being predicted as foam by the model. The overall loss value of the model, which combines the labeled and unlabeled losses, guides the adjustment of model parameters, making the model more accurate in identifying debris on the water surface; L2 loss measures the difference between the sharpened pseudo-labels of unlabeled samples and the model's predicted labels; the smaller the difference, the better. The smaller; This represents the true label probability distribution of labeled samples; The sharpened pseudo-label probability distribution of unlabeled samples; For the predicted categorical variables of the model; θ represents the learnable parameters of the model, such as the weights and biases of the neural network. The model reduces the overall loss L by adjusting θ, ultimately achieving accurate identification of surface debris. H() is the cross-entropy loss function, which measures the difference between the model's predicted label and the true label.

[0068] Finally, based on the final overall loss function, the hyperparameters are adjusted to achieve the optimal recognition model.

[0069] In this embodiment of the application, the computing and control system is further configured as follows: Based on the identification information of multiple consecutive frames and the positioning data provided by the positioning module, a heat map of water surface garbage distribution is generated and dynamically updated. Based on the heat map of surface debris distribution and the environmental information provided by the obstacle avoidance module, the cleaning route is planned and adjusted in real time.

[0070] In the above process, the computing and control system combines the recognition results of multiple consecutive frames with the precise latitude and longitude coordinates provided by the GNSS positioning module at that frame time, and maps them uniformly to the geographic information system. Based on the mapped geographic information system, the system dynamically generates a heat map of surface debris distribution, which visually demonstrates the spatial aggregation of debris. The path planning algorithm uses this heat map as its core input, prioritizing routes to areas with high debris density.

[0071] Meanwhile, the algorithm receives static and dynamic obstacle information detected by the obstacle avoidance module in real time, balances the global clearing target with local obstacle avoidance safety, and calculates and adjusts a safe and efficient optimal navigation path in real time.

[0072] In this embodiment of the application, the heat map of water surface garbage distribution is mapped using a geographic coordinate system, and different colors or numerical gradients are used to characterize the distribution density, category, or cleaning priority of the garbage.

[0073] In the above process, the density of waste distribution on the base map is represented by a color gradient. For example, a gradient from blue to red indicates a transition from a waste-free state to a high-density waste area. Simultaneously, the system can distinguish and mark the identified waste categories using different icons or color blocks, such as triangles representing plastic and circles representing leaves.

[0074] In addition, the system can also combine factors such as the environmental hazard level of waste type and the difficulty of retrieval to calculate a cleanup priority value for different areas, and display it on a heat map in the form of numbers or contour lines, thus providing richer dimensions for intelligent decision-making.

[0075] In this embodiment of the application, the obstacle avoidance module includes a millimeter-wave radar.

[0076] In the aforementioned structure, the radar is installed at the front of the ship's platform. The radar continuously transmits millimeter waves into the surrounding environment and receives echoes. It can accurately detect the distance, relative speed, and azimuth of obstacles within a range of hundreds of meters in front and to the sides. Even in weather conditions with poor visibility, it can provide reliable environmental contours and moving target tracking information for path planning algorithms. It is an indispensable sensor for ensuring navigation safety.

[0077] In this embodiment of the application, the energy and return-to-home system is configured as follows: Real-time monitoring of the remaining power of the battery pack and the loaded capacity of the waste collection bin; When the remaining power is lower than the first preset threshold, or the loaded capacity reaches the second preset threshold, it is determined that a return trip is required. The computing and control system is triggered to plan the optimal path back to the preset base, and the power steering system is controlled to drive the ship platform to return autonomously along the path.

[0078] In the aforementioned system, the energy and return-to-base system continuously monitors the remaining power of the onboard battery pack and the fill rate of the waste collection bins. The system presets two thresholds: a first preset threshold and a second preset threshold. In this application, the first preset threshold is set to 30% (based on battery power), and the second preset threshold is set to 95% (based on waste capacity). When the battery power falls below the first preset threshold, or the waste bin capacity reaches the second preset threshold, the return-to-base condition is triggered. At this time, the calculation and control system immediately pauses the current cleaning task and invokes a path planning algorithm to calculate an optimal path back to a preset dock or charging station, aiming for the lowest energy consumption or shortest time. Subsequently, the system automatically controls the power steering system to drive the ship platform autonomously and safely back along this path.

[0079] In this embodiment of the application, an automatic identification and cleaning method for surface debris is also provided, applied to the aforementioned automatic identification and cleaning device for surface debris, the method comprising: Data on water surface targets are collected using an optical sensing module; The data collected is processed in real time by the surface debris identification model in the computing and control system, and the debris identification information is output. Based on the identification information of multiple consecutive frames and the positioning data of the positioning module, a heat map of the distribution of garbage on the water surface is generated and updated. By integrating the heat map with the environmental perception information from the obstacle avoidance module, a cleaning route is planned. Control the hull platform to navigate along the planned route, and control the garbage collection system to perform salvage operations in areas with dense garbage; The operating status is monitored through the energy and return system, and the control device automatically returns when the return conditions are met.

[0080] In the above process, firstly, the device autonomously cruises, simultaneously collecting environmental spectral data and its own pose through optical sensing and positioning modules. Secondly, the edge computing unit runs a recognition model in real time, parsing waste information from the spectral data. Next, the system integrates spatiotemporal information to generate a dynamic waste heat map and combines obstacle avoidance perception for intelligent navigation planning. Then, the device precisely navigates to the target area and controls the mechanical actuators to complete waste retrieval and collection. Finally, the energy and storage capacity status are monitored throughout the process; once a threshold is reached, a return-to-base procedure is automatically triggered, completing a full automated cleaning task.

[0081] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. An automatic water surface debris identification and cleaning device, characterized in that, include: The hull platform is equipped with a power steering system and a waste collection compartment. A perception system is installed on the hull platform. The perception system includes an optical sensing module, a positioning module, and an obstacle avoidance module. The optical sensing module is used to acquire water surface target data, and the positioning module is used to locate the hull. The computing and control system is equipped with a surface debris recognition model, which processes the data collected by the optical sensing module in real time and outputs debris recognition information. A waste collection and disposal system is installed on the ship's platform. The waste collection and disposal system is controlled by the computing and control system and is used to perform waste retrieval and collection operations. The energy and return-to-home system, including the battery pack and battery management system, is used to monitor the device's power level and waste bin capacity, and to determine and execute the return-to-home decision based on the device's power level and waste bin capacity.

2. The automatic water surface debris identification and cleaning device according to claim 1, characterized in that, The optical sensing module includes a spectral imaging sensor, which is used to acquire reflectance data of water surface targets in multiple continuous or discrete bands to collect raw hyperspectral image data.

3. The automatic water surface debris identification and cleaning device according to claim 2, characterized in that, The execution of the computing and control system includes the following steps: The original hyperspectral image data is preprocessed; Feature band extraction is performed on the preprocessed image; The image after feature extraction is input into the water surface debris recognition model to obtain pixel-level segmentation maps or target-level detection boxes.

4. The automatic identification and cleaning device for surface debris according to claim 3, characterized in that, The feature band extraction employs at least one of principal component analysis, independent component analysis, or linear discriminant analysis.

5. The automatic water surface debris identification and cleaning device according to claim 1, characterized in that, The water surface debris recognition model is trained through the following steps: Obtain hyperspectral image samples from labeled and unlabeled datasets, and generate corresponding labels for the hyperspectral image samples from the labeled datasets; Perform one data augmentation on the labeled dataset with labels, and perform K data augmentations on the unlabeled labeled dataset; For the unlabeled data after K augmentations, the model predicts and calculates the average probability distribution to generate pseudo-labels, and then sharpens the pseudo-labels. The augmented labeled data is mixed with the sharpened unlabeled data to generate mixed data; Calculate the cross-entropy loss of labeled data and the mean squared error loss of unlabeled data, and then sum the two by weight to obtain the overall loss function; The model parameters are optimized based on the overall loss function until the model converges.

6. The automatic water surface debris identification and cleaning device according to claim 1, characterized in that, The computing and control system is further configured to: Based on the identification information of multiple consecutive frames and the positioning data provided by the positioning module, a heat map of the distribution of garbage on the water surface is generated and dynamically updated. Based on the heat map of surface debris distribution and the environmental information provided by the obstacle avoidance module, the cleaning route is planned and adjusted in real time.

7. The automatic identification and cleaning device for surface debris according to claim 6, characterized in that, The heat map of surface waste distribution is mapped using a geographic coordinate system and uses different colors or numerical gradients to characterize the distribution density, type, or cleanup priority of the waste.

8. The automatic identification and cleaning device for surface debris according to claim 1, characterized in that, The obstacle avoidance module includes millimeter-wave radar.

9. The automatic identification and cleaning device for surface debris according to claim 1, characterized in that, The energy and return-to-home system is configured as follows: Real-time monitoring of the remaining power of the battery pack and the loaded capacity of the waste collection bin; When the remaining battery power is lower than the first preset threshold, or the loaded capacity reaches the second preset threshold, it is determined that a return trip is required. The computing and control system is triggered to plan the optimal path back to the preset base, and the power steering system is controlled to drive the ship platform to return autonomously along the path.

10. A method for automatic identification and cleaning of debris on the water surface, characterized in that, The method, applied to the apparatus as described in any one of claims 1 to 9, comprises: Data on water surface targets are collected using an optical sensing module; The collected data is processed in real time by a surface debris identification model in the calculation and control system, and the debris identification information is output. Based on the identification information and positioning data from multiple consecutive frames, a heat map of surface debris distribution is generated and updated. By integrating environmental perception information from heat maps and obstacle avoidance modules, a cleaning route is planned. Control the hull platform to navigate along the planned route and control the garbage collection system to perform salvage operations in areas with dense garbage; The system monitors the operational status through an energy and return-to-home system, and automatically controls the device to return to home when the return-to-home conditions are met.

Images