A method, apparatus, and computing device for controlling a decontamination device

Abstract

The application provides a kind of pollution cleaning equipment control method, device and computing device, it is related to the technical field of water power station pollution cleaning.The specific technical scheme is: the first feature of water body and the second feature of water body surface floating object are acquired;First feature includes the liquid level and flow of water body at trash rack, and second feature includes the coverage of floating object to water body and the density of floating object;Determine the flow obstruction coefficient of water body based on the first feature;Determine the surface density coefficient of floating object based on the second feature;According to flow obstruction coefficient and surface density coefficient, determine the floating object aggregation coefficient of water body;In the case where floating object aggregation coefficient is greater than the preset first aggregation coefficient threshold, control the pollution cleaning equipment to start.The present scheme can improve the accuracy of pollution cleaning equipment control according to the multimodal information of water body.

Description

Technical Field

[0001] This application relates to the field of hydropower station cleaning technology, and in particular to a cleaning equipment control method, device, and computing device. Background Technology

[0002] During the operation of a hydropower station, floating debris (such as branches and aquatic plants) carried by the upstream water flow will accumulate at the water intake of the hydropower station, which will affect the normal operation of the turbine units. Therefore, cleaning equipment is needed to remove the floating debris.

[0003] Currently, in the process of controlling the trash rack, visual recognition technology is first used to identify floating objects in the water at the trash rack, and then the trash rack is controlled to operate and clean the trash. However, visual recognition is easily affected by ambient light and cannot accurately identify the floating objects at the trash rack, which limits the control accuracy of the trash rack and easily leads to problems such as untimely or incorrect cleaning. Summary of the Invention

[0004] This application provides a method, apparatus, and computing device for controlling cleaning equipment, which improves the accuracy of controlling cleaning equipment based on multimodal information of the water body.

[0005] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: Firstly, a method for controlling a cleaning device is provided, the method comprising: First, the system acquires a first characteristic of the water body and a second characteristic of floating debris on its surface. The first characteristic includes the water level and flow rate at the trash rack, while the second characteristic includes the coverage area and density of the floating debris. Next, based on the first characteristic, the system determines the flow obstruction coefficient of the water body; this coefficient characterizes the degree of obstruction caused by the accumulation of floating debris at the trash rack, affecting its flow capacity. Then, based on the second characteristic, the system determines the surface density coefficient of the floating debris; this coefficient characterizes the density of floating debris on the water surface. Then, based on the flow obstruction coefficient and the surface density coefficient, the system determines the floating debris aggregation coefficient of the water body. Finally, if the floating debris aggregation coefficient exceeds a preset first aggregation coefficient threshold, the system activates the cleaning equipment.

[0006] In this scheme, the flow obstruction coefficient is determined by acquiring the liquid level and flow rate at the trash rack, which are the first characteristics of the water body, to characterize the degree of obstruction caused by floating debris accumulation on the flow capacity of the trash rack. Simultaneously, the surface density coefficient is determined by acquiring the coverage and density of floating debris on the water surface, which are the second characteristics of the floating debris, to characterize the density of floating debris on the water surface. The flow obstruction coefficient and the surface density coefficient are then fused to determine the floating debris aggregation coefficient. When the floating debris aggregation coefficient exceeds a preset threshold, the cleaning equipment is activated. This application integrates multimodal information of the water body (e.g., the hydraulic characteristics of the water flow state and visual characteristics reflecting the spatial distribution of floating debris) into the control decision of the cleaning equipment, which can improve the accuracy of floating debris aggregation state identification, thereby improving the accuracy of cleaning equipment control and avoiding problems such as untimely or incorrect cleaning.

[0007] In one possible implementation, controlling the cleaning equipment to start when the floating debris aggregation coefficient is greater than a preset first aggregation coefficient threshold includes: controlling the cleaning equipment to start when a first condition is met and the floating debris aggregation coefficient is greater than the preset first aggregation coefficient threshold. The first condition includes any one of the following: coverage rate greater than a preset first coverage rate threshold; density greater than a preset first density threshold; liquid level difference at the trash rack greater than a preset first liquid level difference threshold; or, flow rate at the trash rack greater than a preset first flow rate attenuation threshold and density greater than a preset second density threshold.

[0008] Thus, by introducing a first condition as a redundancy check in the startup control, and further confirming that at least one original parameter has reached the abnormal threshold based on the fusion result meeting the standard, the risk of false triggering can be effectively reduced and the reliability of startup decision can be improved.

[0009] In another possible implementation, the method further includes: redetermining the floating debris aggregation coefficient during the operation of the cleaning equipment. If the redetermined floating debris aggregation coefficient is less than a second aggregation coefficient threshold and a preset second condition is met, the cleaning equipment is shut down, wherein the second aggregation coefficient threshold is less than a first aggregation coefficient threshold. The second condition includes: coverage rate less than a preset second coverage rate threshold, density less than a third density threshold, liquid level difference at the trash rack less than a preset second liquid level difference threshold, and flow rate attenuation at the trash rack less than a preset second flow rate attenuation threshold.

[0010] Thus, once the cleaning equipment is started, as the cleaning operation progresses, the floating debris accumulated in front of the trash rack is gradually cleared, the liquid level difference gradually decreases, the flow rate gradually recovers, and the coverage and density gradually decrease. Shutdown is triggered when the fusion result falls below the shutdown threshold and all original parameters have returned to normal ranges. This prevents the cleaning equipment from continuing to operate after the floating debris has been cleared to an acceptable level, avoiding idling and over-cleaning, thereby reducing energy consumption and equipment wear.

[0011] In another possible implementation, obtaining a second feature of the floating objects on the water surface includes: acquiring a surface image of the water body; processing the surface image using a visual recognition model to identify the total coverage area of ​​the floating objects and the total number of floating objects; and determining the second feature based on the total coverage area, the total number of floating objects, and the area of ​​the water surface.

[0012] In this way, by processing surface images through a visual recognition model, the total coverage area and total number of floating objects can be determined, thereby determining the coverage rate and density, achieving efficient extraction and accurate quantification of the visual features of floating objects.

[0013] In another possible implementation, the visual recognition model includes a backbone network, a neck network, and a head network. The backbone network includes a multi-dimensional attention convolution module and a multi-scale pooling fusion module. The head network includes a multi-branch reparameterization module. The visual recognition model processes surface images to identify the total coverage area and total number of floating objects. This includes: using the multi-dimensional attention convolution module of the backbone network to extract features from the surface image and apply attention weights to obtain multi-layered weighted feature maps; attention weighting is used to adjust the weights of the feature maps along the channel and spatial dimensions during feature extraction to enhance the feature responses corresponding to the floating objects; and using the backbone network... The network's multi-scale pooling fusion module adaptively pools and convolutionally fuses the deepest feature map in the multi-layer weighted feature map to update the multi-layer weighted feature map. The updated multi-layer weighted feature map is then fused through the neck network to obtain a multi-path fused feature map. The multi-branch reparameterization module of the head network performs feature enhancement processing on the multi-path fused feature map, and object detection and pixel segmentation processing are performed based on the enhanced fused feature map to obtain the detection box and pixel segmentation mask corresponding to each floating object in the surface image. Based on the detection box corresponding to each floating object, the total number of floating objects is determined, and based on the pixel segmentation mask corresponding to each floating object, the total coverage area of ​​the floating objects is determined.

[0014] Thus, by introducing a multi-dimensional attention convolution module into the backbone network, a multi-scale pooling fusion module between the backbone network and the neck network, and a multi-branch reparameterization module into the head network, the model's feature extraction capability, multi-scale perception capability, and pixel-level segmentation accuracy under complex lighting conditions can be enhanced, thereby improving the detection and segmentation performance of the visual recognition model in complex water surface environments.

[0015] In another possible implementation, the flow obstruction coefficient and the surface density coefficient are fused to determine the floating debris aggregation coefficient of the water body. This includes: First, based on preset rules, a first weight value for the flow obstruction coefficient and a second weight value for the surface density coefficient are determined. The sum of the first weight value and the second weight value is 1. Then, based on the first weight value and the second weight value, the flow obstruction coefficient and the surface density coefficient are weighted and summed to determine the floating debris aggregation coefficient of the water body. The preset rules include: if the image quality of the water body's surface image meets the quality requirements, the first weight value is greater than the second weight value; otherwise, the first weight value is less than the second weight value.

[0016] Thus, by dynamically adjusting the weights according to image quality during the fusion process and resetting the corresponding weights to zero when the sensor or vision model fails, the fusion result can adapt to different environmental conditions, while maintaining basic control capabilities even when a single mode fails, thereby improving the robustness and reliability of control decisions.

[0017] Secondly, a cleaning equipment control device is provided, the device comprising: The acquisition module acquires a first feature of the water body and a second feature of floating objects on the water surface. The first feature includes the water level and flow rate at the trash rack, and the second feature includes the coverage of the floating objects on the water body and the density of the floating objects.

[0018] The processing module determines the flow obstruction coefficient of the water body based on a first feature. The flow obstruction coefficient characterizes the degree of blockage caused by floating debris accumulation at the trash rack, affecting the rack's flow capacity. Based on a second feature, it determines the surface density coefficient of the floating debris. The surface density coefficient characterizes the density of floating debris on the water surface. Based on the flow obstruction coefficient and the surface density coefficient, the floating debris aggregation coefficient of the water body is determined. If the floating debris aggregation coefficient exceeds a preset first aggregation coefficient threshold, the cleaning equipment is activated.

[0019] Thirdly, a computing device is provided, comprising: a memory and at least one processor. The memory is communicatively connected to the processor. The memory is used to store computer program code, which includes computer instructions. When the processor executes the computer instructions, it causes the computing device to perform a method as described in the first aspect and any possible implementation thereof.

[0020] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer instructions. When these computer instructions are executed by a processor, they are used to implement the method as described in the first aspect and any possible implementation thereof.

[0021] Fifthly, embodiments of this application provide a computer program product that, when run on a computer or executed by the computer's processor, implements the method described in the first aspect and any possible design thereof. The computer may be a computing device as described in the second aspect and any possible implementation thereof.

[0022] Understandably, the beneficial effects achieved by the cleaning equipment control device of the second aspect, the computing device of the third aspect, the computer-readable storage medium of the fourth aspect, and the computer program product of the fifth aspect provided above can be referred to as the beneficial effects of the first aspect and any possible implementation thereof, which will not be repeated here. Attached Figure Description

[0023] Figure 1 A schematic diagram of the architecture of a cleaning equipment control system provided in an embodiment of this application; Figure 2 This is one of the flowcharts illustrating a method for controlling a cleaning device according to an embodiment of this application; Figure 3 A second schematic flowchart illustrating a method for controlling a cleaning device provided in this application embodiment; Figure 4 A schematic flowchart illustrating a method for shutting down a cleaning device, as provided in an embodiment of this application; Figure 5 A third schematic flowchart illustrating a method for controlling a cleaning device provided in this application embodiment; Figure 6 A flowchart illustrating a process for determining coverage and density, provided for an embodiment of this application; Figure 7 A schematic diagram of the architecture of an improved visual recognition model provided in an embodiment of this application; Figure 8 A schematic diagram of the structure of a multidimensional attention convolution module provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of a multi-scale pooling fusion module provided in an embodiment of this application; Figure 10 This application provides a schematic diagram of the structure of a multi-branch reparameterization module. Figure 11 A schematic diagram illustrating a processing procedure based on an improved visual recognition model, provided for an embodiment of this application; Figure 12 A fourth schematic flowchart illustrating a method for controlling a cleaning device provided in this application embodiment; Figure 13 This is a schematic diagram of the structure of a cleaning equipment control device provided in an embodiment of this application; Figure 14 This is a schematic diagram of the hardware structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0024] Hereinafter, 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 technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0025] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0026] In the technical solutions provided in this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved are all information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision and disclosure of the above information and data all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0027] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0028] In the control of trash racking equipment, visual recognition technology is typically used to identify floating debris in the water at the trash rack, and then the equipment is controlled to clean it. However, visual recognition is easily affected by factors such as ambient light, water surface reflection, and shadows, leading to a decrease in the quality of the acquired images. For example, strong light can cause water surface reflection to obscure the area of ​​floating debris, shadows can change the color and texture of floating debris, and waves can cause partial occlusion or shape changes of floating debris. This makes it difficult for the visual recognition model to accurately detect and segment the complete outline and coverage area of ​​the floating debris, thus making it difficult to accurately identify the actual accumulation of floating debris at the trash rack. Consequently, the control accuracy of the trash racking equipment is limited, and problems such as untimely or incorrect cleaning are prone to occur.

[0029] Therefore, reducing the interference of ambient light, improving the accuracy of identifying the state of floating objects in water, and improving the control accuracy of cleaning equipment are technical problems that urgently need to be solved in this field.

[0030] In view of this, this application provides a method for controlling a cleaning equipment, which uses multimodal information such as the hydraulic characteristics and visual characteristics of the water body to make control decisions on the accumulation state of floating objects in front of the trash rack, so as to reduce the impact of ambient light interference on the identification of floating objects, improve the accuracy of floating object state identification, and thus improve the accuracy of the cleaning equipment control.

[0031] The cleaning equipment control method provided in this application embodiment can be applied to the floating debris cleaning process of hydropower stations. It is suitable for the detection of the floating debris accumulation status at the trash rack at the water inlet of hydropower stations and the linkage control of the cleaning equipment. It can also be applied to other water diversion power stations or pumping stations with trash rack structures.

[0032] The cleaning equipment control method provided in this application embodiment can be applied to a computing device. The computing device can also be an embedded controller integrated into the cleaning equipment. In this case, the computing device is deployed locally on the cleaning equipment and communicates with the actuator of the cleaning equipment through an internal bus.

[0033] In other embodiments, the computing device may be deployed independently of the cleaning equipment and communicate with it. In this case, the computing device may be a standalone server (e.g., a cloud server, edge computing server) or an industrial control computer. The computing device and the cleaning equipment are connected via a wired or wireless communication network. The computing device sends the generated startup command to the controller of the cleaning equipment to control the cleaning equipment to perform the cleaning operation. This application does not limit the specific form of the computing device.

[0034] In an exemplary embodiment, the deployment of a standalone computing device is used as an example for description, such as... Figure 1The diagram shown is an architectural schematic of a cleaning equipment control system provided in an embodiment of this application, including: a computing device 101, a cleaning equipment 102, an edge gateway device 103, a vision sensor 104, a liquid level sensor 105, and a flow sensor 106.

[0035] The computing device 101 is communicatively connected to the cleaning device 102 and the edge gateway device 103. Meanwhile, the edge gateway device 103 is communicatively connected to the vision sensor 104, the liquid level sensor 105, and the flow sensor 106.

[0036] The visual sensor 104 is used to acquire surface image data of the water body.

[0037] like Figure 1 As shown, the visual sensor 104 is deployed on the upper side of the trash rack. The trash rack is a grid-like structure arranged in the water body to intercept floating debris carried in by the water flow, preventing it from entering the turbine unit. The water body refers to the water flow within the area of ​​the intake channel (or forebay or inlet) in front of the trash rack, such as... Figure 1 As shown, the left side of the trash rack is the water body, and the right side of the trash rack is the water inlet of the water turbine.

[0038] For example, the visual sensor 104 includes a visible light camera and an infrared night vision supplementary light camera, etc.

[0039] In some embodiments, multiple visual sensors 104 may be included, for example, multiple visual sensors 104 may be deployed in the middle, at both ends, or above the water body on the upper side of the trash rack.

[0040] The liquid level sensor 105 is used to collect liquid level data at the trash rack.

[0041] like Figure 1 As shown, the liquid level sensor 105 includes multiple sensors, which can be deployed in front of and behind the trash rack, for example. Figure 1 The left and right sides are shown in the middle, representing the planar direction.

[0042] In some embodiments, the level sensor 105 is an ultrasonic level sensor, which may also be deployed above the trash rack.

[0043] Flow sensor 106 is used to collect flow data at the trash rack.

[0044] like Figure 1 As shown, the flow sensor 106 is deployed on the rear side of the trash rack. Figure 1 (Displayed on the right).

[0045] For example, the flow sensor 106 may be a Doppler flow meter or an ultrasonic flow meter.

[0046] Edge gateway device 103 is used to receive surface images collected by vision sensor 104, liquid level data collected by liquid level sensor 105, and flow data collected by flow sensor 106, and after performing time-series processing on the received data, obtain liquid level, flow rate and surface images of water body, and send them to computing device 101.

[0047] For example, the edge gateway device 103 is deployed in a cabinet near the water body and communicates with each sensor via Ethernet or Controller Area Network (CAN) bus.

[0048] In one possible implementation, the edge gateway device 103 is equipped with a clock synchronization device. After acquiring liquid level data, flow data, and surface images, the device performs clock synchronization processing on the acquired multi-source data and sends the synchronized data to the computing device 101.

[0049] For example, the edge gateway acts as the core of the master clock, internally comprising a clock synchronization module, a hardware triggering module, and a synchronization processing module. The clock synchronization module obtains a globally unified reference time through the Global Positioning System (GPS), Precision Time Protocol (PTP), or Network Time Protocol (NTP). The hardware triggering module outputs a synchronization pulse signal based on this reference time and sends it to the smart camera, liquid level sensor, and flow sensor; the synchronization processing module is used to append timestamps to the collected data.

[0050] After receiving the synchronization pulse signal, the vision sensor 104, the liquid level sensor 105, and the flow sensor 106 synchronously perform data acquisition and output surface images, liquid level data, and flow data with timestamps. The above multi-source data is transmitted to the edge gateway device 103 via the bus. The edge gateway device 103 performs buffering and completion processing on the multi-source data, completes the timing alignment of the multi-source data, and finally sends the timing-synchronized surface image, liquid level data, and flow data to the computing device.

[0051] The computing device 101 is used to acquire a first feature of the water body and a second feature of floating debris on the water surface based on liquid level data, flow rate data, and surface images. The first feature includes the liquid level and flow rate at the trash rack, and the second feature includes the coverage rate and density of floating debris on the water body. Based on the first feature, a flow obstruction coefficient is determined; the flow obstruction coefficient characterizes the degree of obstruction caused by the accumulation of floating debris at the trash rack, affecting the flow capacity of the trash rack. Based on the second feature, a surface density coefficient is determined; the surface density coefficient characterizes the density of floating debris on the water surface. The flow obstruction coefficient and the surface density coefficient are fused to determine the floating debris aggregation coefficient of the water body. If the floating debris aggregation coefficient is greater than a preset first aggregation coefficient threshold, the cleaning device 102 is activated.

[0052] The embodiments of this application are described below.

[0053] In an exemplary embodiment, when the cleaning equipment control method provided in this application is applied to the aforementioned computing device, such as... Figure 2 As shown, the specific steps may include S101-S105: S101. Obtain the first feature of the water body and the second feature of the floating objects on the surface of the water body.

[0054] The first feature includes the water level and flow rate at the trash rack, and the second feature includes the coverage of the floating debris on the water body and the density of the floating debris.

[0055] The water level at the trash rack refers to the water height at the trash rack (behind the trash rack). A lower water level indicates more severe accumulation of floating debris. The water flow rate at the trash rack refers to the actual flow rate through the cross-section of the trash rack. A lower flow rate indicates more severe accumulation of floating debris. The water level reflects the water level change caused by accumulation, while the flow rate reflects the reduction in flow due to accumulation. Both reflect the degree of flow obstruction from different perspectives.

[0056] The surface coverage of a water body refers to the proportion of the total water area covered by floating debris, reflecting the breadth of the floating debris's distribution on the water surface. The density of floating debris on the water surface refers to the number of individual floating objects per unit area, reflecting the degree of aggregation and compactness of the floating debris on the water surface.

[0057] By determining the multimodal information belonging to the hydraulic and visual dimensions, the accumulation status of floating objects in front of the trash rack can be characterized from both fluid dynamics and spatial distribution perspectives. This achieves complementary enhancement of hydraulic and visual features, thereby improving the comprehensiveness of the representation of the floating object accumulation status and reducing the identification error of the floating object accumulation status in the water body.

[0058] In some embodiments, the computing device can acquire liquid level data, flow rate data, and surface image data collected by sensors from an edge gateway device. Each type of data includes measured values ​​at different acquisition times. The computing device then processes the liquid level data, flow rate data, and surface image data to obtain a first feature and a second feature.

[0059] The above processing includes: representative value selection, standardization processing, and visual recognition processing.

[0060] Representative value selection means selecting values ​​from different acquisition times that are representative of the data acquisition time as the values ​​to be processed by the computing device. For example, among the five acquisition times from t1 to t5, if the water state data at time t3 changes more gradually than the water state data at times t2 and t4, then the value corresponding to time t3 is selected to obtain the first feature and the second feature.

[0061] Standardization processing involves processing the parameters of different dimensions in the first and second features mentioned above, mapping their values ​​to the range of [0,1].

[0062] Visual recognition processing, also known as the recognition processing of surface images, involves obtaining second features. This process can be referred to below. Figure 6 And related descriptions, which will not be elaborated here.

[0063] In one possible implementation, for the liquid level, the computing device can perform linear standardization processing on the liquid level based on preset maximum and minimum reference liquid levels.

[0064] For example, the process satisfies the following relationship:

[0065] in, This indicates the liquid level collected by the sensor. This indicates the minimum reference liquid level under preset normal operating conditions. This indicates the maximum reference liquid level under preset normal operating conditions, and L indicates the liquid level after processing.

[0066] Indicates will Cut off to the [0,1] interval (i.e. When less than 0, take 0. If the value is greater than 1, take 1. Take the value between 0 and 1. itself).

[0067] That is, when At this point, L can be set to 0. When At this point, L can be set to 1. It can be understood that when the measured liquid level is lower than the lower limit, it indicates that the accumulation of floating objects has caused severe obstruction of the flow, and the liquid level drops beyond the normal fluctuation range. At this time, L takes the minimum value of 0. When the measured liquid level is higher than the upper limit, it indicates that the flow is smooth and the liquid level is above the upper limit of the normal fluctuation. At this time, L takes the maximum value of 1.

[0068] For example, assuming the liquid level fluctuates between 5m and 10m under normal operating conditions, that is... =5m, =10m. If the actual liquid level is... =4.5m, since 4.5 < 5, that is < At this point, set L to 0. If the measured liquid level... =7m, then L=(7-5) / (10-5)=2 / 5=0.4. If the measured liquid level is... =11m, since 11>10, that is > At this point, L is set to 1.

[0069] In one possible implementation, for traffic, the computing device can perform linear standardization processing on the traffic based on a preset reference traffic and in combination with the allowable fluctuation range of the traffic.

[0070] For example, the process satisfies the following relationship:

[0071] in, This indicates the flow rate collected by the sensor. This indicates the preset reference flow rate. This represents the lower limit coefficient for permissible traffic fluctuations. This represents the allowable upper limit coefficient for traffic fluctuation, where F represents the processed traffic. and Satisfy: 0 < <1< .

[0072] Indicates will Cut off to the [0,1] interval (i.e. When less than 0, take 0. If the value is greater than 1, take 1. Take the value between 0 and 1. itself).

[0073] That is, when < At this time, you can set =0; when > At that time, you can set The value of 1 indicates that when the measured flow rate is below the lower limit, it means that the flow through the trash rack is severely obstructed and the flow rate has exceeded the normal fluctuation range. In this case, the standardized value is the minimum value of 0. When the measured flow rate is above the upper limit, it means that the flow is smooth and the flow rate is above the upper limit of the normal fluctuation range. In this case, the standardized value is the maximum value of 1.

[0074] For example, It can be 0.9. The value is 1.1, assuming a reference flow rate of 100 cubic meters per second. Therefore, the lower limit of allowable flow fluctuation is 0.9 × 100 = 90. The maximum allowable flow fluctuation is 1.1 × 100 = 110. If the collected traffic is 85 Then set F to 0. If the collected traffic is 92... Substituting into equation (2), F is set to (92-90) / (110-90) = 0.1. If the collected flow rate is 115... Since 115 > 110, F is set to 1.

[0075] In one possible implementation, the computing device can linearly normalize the density based on a preset first density threshold.

[0076] For example, the process satisfies the following relationship:

[0077] in, This indicates the density of floating objects on the water surface identified after visual recognition processing. This represents the preset first density threshold (i.e., the maximum tolerable density). This represents the minimum value function, used to... The result is truncated to no more than 1.

[0078] For example, suppose For 5 / ,like For 1 / If, then D = 0.2. For 6 / If so, then D=1.

[0079] S102. Based on the first feature, determine the overcurrent blocking coefficient.

[0080] The flow blockage coefficient characterizes the degree of blockage caused by floating debris accumulating at the trash rack, thus affecting the trash rack's flow capacity.

[0081] Liquid level and flow rate belong to the hydraulic dimension, and the two are parameters that change in tandem: when floating debris accumulates at the trash rack, the flow area decreases, causing the liquid level behind the trash rack to drop, and at the same time, the flow rate through the trash rack section also decreases.

[0082] When floating debris accumulates at the trash rack, it simultaneously affects both the liquid level and flow rate at the rack. Specifically, the liquid level decreases and the flow rate decreases. These changes reflect the impact of debris accumulation on the flow from different perspectives. Therefore, determining the flow blockage coefficient by comprehensively considering both liquid level and flow rate can improve the accuracy of assessing the flow blockage status of the trash rack.

[0083] In some embodiments, based on the above-described standardized processing results of liquid level and flow rate, the computing device can perform a weighted summation of the two to obtain the overflow blockage coefficient.

[0084] In one possible implementation, the computing device has a preset weight value corresponding to the liquid level. Weight value corresponding to traffic The weighted sum of the standardized liquid level L and flow rate F is obtained. The weighted sum represents the overall unobstructed flow rate after considering both liquid level and flow rate (a higher value indicates better flow). Since the flow blockage coefficient represents the degree of blockage (a higher value indicates greater blockage), the calculation device subtracts this weighted sum from 1 to obtain the flow blockage coefficient S. This process satisfies the following relationship:

[0085] Where S represents the overcurrent blocking coefficient, This represents the weight value corresponding to the liquid level. This represents the weight value corresponding to the traffic. The sum of is 1.

[0086] For example, suppose It is 0.5. Given 0.5, L is 0.4, and F is 0.75, substituting these values ​​into equation (4) yields S = 0.425.

[0087] In addition, the weight values ​​corresponding to liquid level and flow rate can be dynamically adjusted according to operating conditions to adapt to different operating environments.

[0088] In one possible implementation, when the flow rate decreases by more than a preset rate of change threshold per unit time, it is determined that the upstream water flow is fluctuating significantly. At this time, the flow data is significantly affected by the fluctuation of the water flow, and the reliability is reduced. Therefore, the weight value of the liquid level can be increased.

[0089] In one possible implementation, when the drop in liquid level is less than a preset liquid level change threshold and the drop in flow rate is greater than a preset flow rate change threshold, the degree of blockage of the trash rack is determined to be relatively mild. At this time, the liquid level change is not significant and the flow rate change can better reflect the change in the flow state, so the weight value of the flow rate can be increased.

[0090] In other embodiments, the computing device uses a distance fusion method to determine the flow blockage coefficient. The computing device uses the normalized liquid level value L and flow rate value F as coordinate points (L, F) on a two-dimensional plane, calculates the Euclidean distance between this point and the unobstructed state reference point (1,1), and uses this distance as the flow blockage coefficient. This process satisfies the following relationship:

[0091] Understandably, when the trash rack is unobstructed, both L and F approach 1, which is close to the reference point (1,1), and S approaches 0. When floating debris accumulates and obstructs the flow, both L and F decrease, the point deviates from the reference point, and S increases. Therefore, S can continuously reflect the degree of flow obstruction, and a larger S value indicates a more severe obstruction, which facilitates subsequent integration with the surface density coefficient.

[0092] S103. Based on the second feature, determine the surface density coefficient.

[0093] Among them, the surface density coefficient characterizes the density of floating objects on the water surface.

[0094] Coverage and density are both visual parameters, reflecting the aggregation state of floating debris on the water surface from different perspectives. Relying solely on coverage, large areas of single-layer floating debris (such as duckweed) have high coverage but few individual particles, limiting their actual impact on debris screen blockages but tending to control debris removal. Relying solely on density, a large number of dispersed individuals have high density but limited coverage, making it impossible to comprehensively assess the degree of accumulation. Therefore, combining coverage and density provides a more comprehensive and accurate reflection of the aggregation state of floating debris on the water surface.

[0095] In some embodiments, the computing device may perform a weighted summation of coverage and density to obtain an overcurrent blocking coefficient. Exemplarily, this process satisfies the following relationship:

[0096] Where C represents coverage and D represents density. This represents the weight value corresponding to the coverage rate. This represents the weight value corresponding to the density. The sum of is 1.

[0097] For example, suppose It is 0.4. Since C is 0.6, D is 0.8, and D is 0.3, substituting these values ​​into equation (6), we get V = 0.50.

[0098] S104. Based on the fusion processing of the flow blockage coefficient and the surface density coefficient, determine the floating object aggregation coefficient of the water body.

[0099] Among them, the floating debris accumulation coefficient characterizes the overall severity of floating debris accumulation in front of the trash rack. The flow blockage coefficient reflects the impact of floating debris accumulation on flow capacity from a hydraulic perspective, while the surface density coefficient reflects the aggregation state of floating debris on the water surface from a visual perspective. Both represent different aspects of the floating debris accumulation state from different dimensions. Integrating the two allows for the comprehensive use of hydraulic and visual information to jointly assess the overall severity of floating debris accumulation from multiple dimensions, improving the comprehensiveness and accuracy of the assessment.

[0100] In one possible implementation, the computing device performs a weighted summation of the overcurrent blockage coefficient and the surface density coefficient to obtain the floating debris aggregation coefficient. This process is described below. Figure 12 And related descriptions, which will not be elaborated here.

[0101] In another possible implementation, the computing device can determine the floating object aggregation coefficient by taking the maximum value. Specifically, the computing device compares the flow blockage coefficient and the surface density coefficient, and uses the larger of the two as the floating object aggregation coefficient. Understandably, taking the maximum value can avoid lowering the fusion result due to a low coefficient in one dimension, thereby improving the response sensitivity to the floating object accumulation state.

[0102] In another possible implementation, the computing device can also use a primary-secondary adjustment method to determine the floating debris aggregation coefficient. The computing device uses either the overflow blockage coefficient or the surface density coefficient as the primary decision coefficient, and the other as the secondary decision coefficient. The primary decision coefficient is adjusted based on the value of the secondary decision coefficient, and the adjusted result is used as the floating debris aggregation coefficient.

[0103] For example, the process satisfies the following relationship:

[0104] Where G represents the floating object aggregation coefficient, S is the primary determination coefficient, and V is the secondary determination coefficient.

[0105] In equation (7), when both S and V are high, their multiplication amplifies the aggregation effect, making G more prominent; when S is high but V is low, it indicates that the hydraulic dimension shows blockage while the visual dimension is not confirmed, the adjustment range is limited, and the value of G is relatively conservative. It is understandable that this method uses hydraulic data as the main source and visual data as a supplement, which can avoid control deviations caused by misjudgment of a single dimension and improve the accuracy of determining the floating object aggregation coefficient.

[0106] Alternatively, the surface density coefficient V can be used as the primary decision coefficient, and the overcurrent blockage coefficient S as the secondary decision coefficient. The adjustment logic is similar and will not be elaborated here.

[0107] S105. If the floating debris aggregation coefficient is greater than the preset first aggregation coefficient threshold, control the cleaning equipment to start.

[0108] The preset first aggregation coefficient threshold is used to determine whether the accumulation of floating debris in front of the trash rack has reached a level that requires the activation of the cleaning equipment. When the floating debris aggregation coefficient is greater than this threshold, it indicates that the accumulation of floating debris is quite serious and poses an impact or potential threat to the normal operation of the turbine unit. At this time, it is necessary to activate the cleaning equipment for cleaning.

[0109] For example, the first aggregation coefficient threshold may be 0.7, 0.8 or 0.9, etc., and the embodiments of this application do not limit the first aggregation coefficient threshold.

[0110] In one possible implementation, the computing device sends a start command to the controller of the cleaning equipment. After receiving the start command, the controller of the cleaning equipment controls the drive motor of the cleaning equipment to run, so that the cleaning equipment can perform cleaning operations.

[0111] In some embodiments, in addition to controlling the start of the cleaning equipment when the floating object aggregation coefficient is greater than a preset first aggregation coefficient threshold, the computing device can also set more aggregation coefficient thresholds to classify and evaluate the floating object aggregation state and push corresponding status prompts to the staff so that the staff can understand the on-site situation in a timely manner.

[0112] For example, the computing device may set a third clustering coefficient threshold, wherein the first clustering coefficient threshold > the third clustering coefficient threshold > the second clustering coefficient. For example, the first clustering coefficient threshold is 0.8, the third clustering coefficient is 0.5, and the second clustering coefficient is 0.2.

[0113] When the floating debris aggregation coefficient falls into different threshold ranges, it indicates different levels of floating debris aggregation: When the floating object aggregation coefficient is greater than or equal to the third aggregation coefficient threshold and less than the first aggregation coefficient threshold, the floating object aggregation state is determined to be abnormal. The computing device triggers an abnormal alarm and pushes it to the staff, and continuously monitors the changing trend of the floating object aggregation coefficient. If the floating object aggregation coefficient subsequently rises to the first aggregation coefficient threshold, the control is activated.

[0114] When the floating object aggregation coefficient is greater than or equal to the second aggregation coefficient threshold and less than the third aggregation coefficient threshold, the floating object aggregation state is determined to be an early warning, indicating that there is a slight abnormality on site but the activation conditions have not yet been met. The calculation equipment triggers the early warning information to remind the staff to pay more attention.

[0115] When the floating debris aggregation coefficient is less than the second aggregation coefficient threshold, the floating debris aggregation state is deemed qualified, indicating that the on-site operation is normal and the cleaning equipment is kept in standby mode.

[0116] Thus, by acquiring the liquid level and flow rate at the trash rack, the flow obstruction coefficient is determined to characterize the degree of obstruction caused by floating debris accumulation to the trash rack's flow capacity. Simultaneously, by determining the coverage and density of floating debris on the water surface, a surface density coefficient is determined to characterize the density of floating debris on the water surface. The flow obstruction coefficient and surface density coefficient are then fused to determine the floating debris aggregation coefficient. When the floating debris aggregation coefficient exceeds a preset threshold, the cleaning equipment is activated. This application integrates multimodal information about the water body (e.g., the hydraulic characteristics of the water flow state and visual characteristics reflecting the spatial distribution of floating debris) into the control decision of the cleaning equipment. This improves the accuracy of determining the floating debris aggregation state, thereby enhancing the accuracy of the cleaning equipment control and avoiding problems such as untimely or incorrect cleaning.

[0117] The following is an introduction to controlling the start-up of the cleaning equipment.

[0118] In some embodiments, the first feature may further include: the liquid level at the trash rack and the liquid level difference at the trash rack, wherein the liquid level difference at the trash rack refers to the height difference between the liquid levels on the front and rear sides of the trash rack. A larger liquid level difference indicates a more severe blockage of the flow caused by floating debris accumulation. The sensors at the trash rack can be deployed on the front and rear sides of the trash rack respectively, thereby acquiring the liquid levels on the front and rear sides of the trash rack and obtaining the liquid level difference.

[0119] Based on this, when controlling the start-up of the cleaning equipment in S105, the computing device can also set original parameter threshold conditions based on the first feature and the second feature to increase the safety redundancy control of the cleaning equipment, such as... Figure 3 As shown, S105 can be specifically implemented as S105A: S105A. When the first condition is met and the floating object aggregation coefficient is greater than the preset first aggregation coefficient threshold, the cleaning equipment is started.

[0120] The floating debris aggregation coefficient is a comprehensive result obtained from the fusion of multimodal information. To further improve the reliability of the startup decision, based on the floating debris aggregation coefficient being greater than a preset first aggregation coefficient threshold, a further judgment is made as to whether a first condition is met, confirming that at least one original parameter has reached a clear anomaly threshold. In other words, startup is only controlled when the fusion result meets the standard and at least one original parameter is clearly abnormal; if the fusion result meets the standard but none of the original parameters have reached the anomaly threshold, the current operating condition is considered to be within the normal fluctuation range of each dimension parameter, and startup is not controlled. By setting this redundant verification condition, the reliability of the startup decision of the cleaning equipment can be effectively improved, and the risk of erroneous control can be reduced.

[0121] In some embodiments, the first condition includes any one of the following: the coverage rate is greater than a preset first coverage rate threshold, the density is greater than a preset first density threshold, the liquid level difference at the trash rack is greater than a preset first liquid level difference threshold, or the flow rate at the trash rack is greater than a preset first flow rate attenuation threshold and the density is greater than a preset second density threshold.

[0122] The liquid level difference at the trash rack refers to the difference in water level height between the front and rear sides of the trash rack; a larger liquid level difference indicates more severe accumulation. The flow rate at the trash rack refers to the change in flow rate through the trash rack cross-section relative to the reference flow rate under normal operating conditions; a larger flow rate attenuation indicates more severe accumulation.

[0123] In some embodiments, the liquid level difference at the trash rack can be obtained by collecting liquid level data from liquid level sensors deployed on the front and rear sides of the trash rack and calculating the difference. The flow rate at the trash rack can be obtained by collecting the current flow rate value from a flow sensor deployed on the rear side of the trash rack and calculating the rate of change based on a preset normal operating condition reference flow rate.

[0124] For example, the traffic attenuation rate can be expressed as: Traffic attenuation rate = (Reference traffic - Current traffic) / Reference traffic × 100%.

[0125] For example, the first coverage threshold can be 25%, 30%, or 35%, etc., and the first density threshold can be 5 units / Or 6 / The first liquid level difference threshold can be 0.25m, 0.3m, or 0.35m, etc.; the first flow rate attenuation rate threshold can be 8%, 10%, or 12%, etc.; and the second density threshold can be 3 units / ... Or 4 / wait.

[0126] For example, the preset first clustering coefficient threshold is 0.7, the first coverage threshold is 30%, and the first density threshold is 5 units / year. The first liquid level difference threshold is 0.3m, the first flow rate threshold is 0.8m³ / s, and the second density threshold is 2 units / second. .

[0127] At a certain moment, the sensor collected the following raw data: the liquid level in front of the trash rack is 0.55m, the liquid level behind it is 0.25m, the level difference is 0.30m, and the current flow rate is 0.75. The coverage rate obtained by visual recognition was 45%, and the density was 7 units / day. .

[0128] According to the standardized processing procedure in S101, the liquid level, flow rate, and density are normalized: Assumption It is 0.5m. It is 0.8m. Given a radius of 0.25m, substituting this into equation (1), we get L = clamp((0.25-0.5) / (0.8-0.5), 0, 1) = 0. Assuming a is 0.9 and b is 1.1, the reference flow rate is... The lower limit is 1.0 m³ / s, and the lower limit is a. It is 0.9 m³ / s, and the upper limit is b. The measured flow rate is 1.1 m³ / s. The value is 0.75 m³ / s. Substituting this into equation (2): F = 0.

[0129] Based on the first density threshold and the density obtained by visual recognition, substitute into equation (3): D=1.

[0130] Assumption , All are 0.5. Calculate the overcurrent blocking coefficient: S=1-(0.5×0+0.5×0)=1.0.

[0131] Assumption , The surface density coefficient is calculated as follows: V = 0.5 × 0.45 + 0.5 × 1.0 = 0.225 + 0.5 = 0.725.

[0132] Calculate the floating debris aggregation coefficient: G = 0.5 × S + 0.5 × V = 0.5 × 1.0 + 0.5 × 0.725 = 0.8625.

[0133] The value of 0.8625 is greater than the preset first aggregation coefficient threshold of 0.7 and meets the first condition (e.g., the liquid level difference is equal to the first liquid level difference threshold). Therefore, the computing device controls the cleaning equipment to start.

[0134] Thus, by introducing a first condition as a redundancy check in the startup control, and further confirming that at least one original parameter has reached the abnormal threshold based on the fusion result meeting the standard, the risk of false triggering can be effectively reduced and the reliability of startup decision can be improved.

[0135] The following section introduces how to control the shutdown of cleaning equipment.

[0136] In some embodiments, after the cleaning equipment is started and running, as the cleaning operation proceeds, the floating debris in front of the trash rack is gradually cleared, and the various state parameters (liquid level difference, flow rate attenuation, coverage, density) gradually return to normal range. At this point, the floating debris aggregation coefficient can be determined, and it can be determined whether the cleaning equipment needs to be shut down.

[0137] In this case, such as Figure 4 As shown, after S105 above, the method further includes steps S106-S107: S106. During the operation of the cleaning equipment, the floating debris aggregation coefficient is re-determined.

[0138] In this step, the computing device periodically acquires the liquid level, flow rate, coverage and density at preset time intervals during the operation of the cleaning equipment, and redetermines the current floating matter aggregation coefficient in the same way as in S101-S104.

[0139] Understandably, re-determining the floating debris aggregation coefficient will also require re-determining the aforementioned liquid level, flow rate, coverage, density, as well as the liquid level difference and flow rate attenuation.

[0140] S107. If the re-determined floating object aggregation coefficient is less than the second aggregation coefficient threshold and the preset second condition is met, control the cleaning equipment to stop.

[0141] The second aggregation coefficient threshold is lower than the first aggregation coefficient threshold. That is, when the aggregation coefficient of the floating objects decreases to below the second aggregation coefficient threshold, it indicates that the degree of floating object accumulation has been significantly reduced, and the fusion criterion has met the conditions for shutdown. For example, the second aggregation coefficient threshold can be 0.2, 0.3, or 0.4, etc., and the value of the second aggregation coefficient threshold is not limited in this application embodiment.

[0142] Based on the same concept as S105A, after the fusion result indicates that shutdown conditions are met, a second condition is further determined to confirm that the original parameters of each dimension have returned to normal ranges, avoiding premature shutdown due to fluctuations in parameters during the fusion process. In other words, shutdown is only triggered when the fusion result is below the shutdown threshold and the original parameters of each dimension have reached the threshold corresponding to the clean state. By setting this redundant verification condition, the reliability of the shutdown decision of the cleaning equipment can be effectively improved, preventing the cleaning equipment from shutting down prematurely before floating debris is completely removed, and ensuring the thoroughness of the cleaning operation.

[0143] In some embodiments, the second condition includes: the coverage rate is less than a preset second coverage rate threshold, the density is less than a third density threshold, the liquid level difference at the trash rack is less than a preset second liquid level difference threshold, and the flow rate at the trash rack is less than a preset second flow rate attenuation threshold.

[0144] For example, the second coverage threshold can be 4%, 5%, or 6%, and the third density threshold can be 1 / Or 2 / The second liquid level difference threshold can be 0.1m or 0.15m, and the second flow rate attenuation threshold can be 4% or 5%.

[0145] For example, the preset second clustering coefficient threshold is 0.3, the second coverage threshold is 5%, and the third density threshold is 1 per [unit]. The second liquid level difference threshold is 0.1m, and the second flow rate attenuation threshold is 5%. The preset normal liquid level range is 0.5m~0.8m, and the preset reference flow rate is 1.0m³ / s.

[0146] At a certain moment, the cleaning equipment is in operation, and the sensors collect the following raw data: the liquid level in front of the trash rack is 0.80m, the liquid level behind it is 0.75m, the level difference is 0.05m, the current flow rate is 0.97m³ / s, the coverage rate is 2%, and the density is 0.5 units / ... All original parameters have met the second condition (liquid level difference 0.05m < 0.1m, flow rate attenuation 3% < 5%, coverage 2% < 5%, density 0.5 units / ). <1 / ).

[0147] According to the standardized processing procedure in S101, the liquid level, flow rate, and density are normalized: Liquid level normalization: The liquid level behind the trash rack is used for normalization. It is 0.5m. It is 0.8m. The value is 0.75m. Substituting into equation (1): L=clamp(0.25 / 0.3,0,1)≈0.833.

[0148] Traffic normalization: a = 0.9, b = 1.1, reference traffic. The measured flow rate is 1.0 m³ / s. The value is 0.97 m³ / s. Substituting into equation (2): F = clamp(0.07 / 0.2, 0, 1) = 0.35.

[0149] Density normalization: The density obtained from visual recognition is 0.5 units / Substituting into equation (3): D = min(0.5 / 5, 1) = 0.1.

[0150] Assumption , The values ​​are all 0.5. The overcurrent blocking coefficient is calculated as: S = 1 - (0.5 × 0.833 + 0.5 × 0.35) = 0.4085.

[0151] Assumption , The surface density coefficient is calculated as follows: V = 0.5 × 0.02 + 0.5 × 0.1 = 0.01 + 0.05 = 0.06.

[0152] Calculate the floating debris aggregation coefficient: G = 0.5 × S + 0.5 × V = 0.20425 + 0.03 = 0.23425.

[0153] The G value of 0.234 is less than the preset second aggregation coefficient threshold of 0.3, and all original parameters have met the second condition (liquid level difference 0.05m < 0.1m, flow rate attenuation 3% < 5%, coverage 2% < 5%, density 0.5 / ). <1 / Therefore, the computing device controls the shutdown of the cleaning equipment.

[0154] Thus, once the cleaning equipment is started, as the cleaning operation progresses, the floating debris accumulated in front of the trash rack is gradually cleared, the liquid level difference gradually decreases, the flow rate gradually recovers, and the coverage and density gradually decrease. Shutdown is triggered when the fusion result falls below the shutdown threshold and all original parameters have returned to normal ranges. This prevents the cleaning equipment from continuing to operate after the floating debris has been cleared to an acceptable level, avoiding idling and over-cleaning, thereby reducing energy consumption and equipment wear.

[0155] In an exemplary embodiment, such as Figure 5 As shown, the control flow of the computing device includes: When the cleaning equipment is not started: 1. Obtain the first feature and the second feature.

[0156] 2. Determine the floating object aggregation coefficient based on the first and second features.

[0157] 3. Whether the floating object aggregation coefficient is greater than the preset first aggregation coefficient.

[0158] If it is greater than, proceed to step 4. If it is less than or equal to, return to step 1.

[0159] 4. Determine that the first condition is met.

[0160] If satisfied, proceed to step 5. If not satisfied, return to step 1.

[0161] 5. Control the start-up of the cleaning equipment.

[0162] During the operation of the cleaning equipment: 6. Redetermine the floating object aggregation coefficient.

[0163] 7. Re-determine whether the floating object aggregation coefficient is less than the preset second aggregation coefficient.

[0164] If it is less than, proceed to step 8; if it is greater than or equal to, return to step 6.

[0165] 8. Determine that the second condition is met.

[0166] If the conditions are met, the cleaning equipment will be shut down. If not, return to step 6.

[0167] 9. Control the shutdown of cleaning equipment.

[0168] The following section introduces the visual recognition process for coverage and density.

[0169] In some embodiments, a computing device can determine the coverage and density of floating debris on the water surface by visually recognizing images of the water surface. For example... Figure 6 As shown, the process of determining the coverage and density of floating debris on the water surface in S101 specifically includes the following steps S1011-S1013: S1011. Obtain a surface image of the water body.

[0170] Combination Figure 1 In addition to the above-mentioned S101, the surface image of the water body is image data that is collected by the visual sensor 104, processed by the edge gateway device 103, and then sent to the computing device 101.

[0171] In some embodiments, the surface image includes a top view or a side view of the water body. The top view is an image taken from above, and the side view is an image taken horizontally from the front or side of the trash rack.

[0172] S1012. Process surface images using a visual recognition model to identify the total coverage area of ​​floating objects and the total number of floating objects.

[0173] The computing device inputs the surface image into a pre-trained visual recognition model for processing. This model can detect and segment floating objects in the surface image, thereby identifying all floating object instances and calculating their total coverage area and total number.

[0174] In some embodiments, the model can be deployed in a computing device, or it can be deployed on an external device (e.g., a cloud server) that is connected to the computing device. The computing device sends the surface image to the external device for processing through an application programming interface (API) or other means, and obtains the total coverage area and the total number from the return result of the API.

[0175] For example, the visual recognition model can be a YOLOv8-Seg model based on object detection algorithms or a masked region convolutional neural network model (Mask R-CNN), etc. The embodiments of this application do not limit the type of visual recognition model.

[0176] In one possible implementation, the visual recognition model employs the YOLOv8-Seg model. The YOLOv8-Seg structure comprises three parts: a backbone network, a neck network, and a head network. Its data processing flow is as follows: the input surface image, after preprocessing (such as scaling and normalization), is input into the backbone network for multi-level feature extraction, outputting multi-level feature maps at different scales. These feature maps at different scales can capture the features of floating objects of different sizes—shallow feature maps have high spatial resolution and are suitable for detecting small floating objects; deep feature maps have a larger receptive field and stronger semantic information, making them suitable for detecting large floating objects.

[0177] Next, the neck network performs multi-scale pooling on the feature map output from the last layer of the backbone network through the SPPF module, and performs cross-scale fusion and enhancement on feature maps of different scales through the FPN+PAN structure to obtain the fused feature map.

[0178] SPPF stands for Spatial Pyramid Pooling Fast, a module in YOLO-type models used for multi-scale feature extraction. It captures contextual information at different scales through sequential pooling operations. FPN stands for Feature Pyramid Network, which transmits deep semantic information to shallow layers through a top-down path, enhancing the semantic expressiveness of feature maps at each scale. PAN stands for Path Aggregation Network, which adds a bottom-up path to FPN, transmitting shallow positional information to deeper layers and achieving bidirectional cross-scale fusion.

[0179] Then, the head network performs target detection and pixel-level segmentation based on the fused feature map to obtain preliminary detection boxes and pixel segmentation masks. The detection boxes are then filtered by non-maximum suppression (NMS) to remove redundant and duplicate detection boxes, and finally output the detection boxes and pixel segmentation masks of floating objects in the surface image.

[0180] Finally, the computing device determines the total coverage area of ​​the floating objects based on the pixel segmentation mask of each floating object, and determines the total number of floating objects based on the detection box of each floating object.

[0181] In one possible implementation, the computing device counts the number of detection frames for each floating object and uses this number as the total number of floating objects.

[0182] In one possible implementation, for the total coverage area of ​​the floating objects, the computing device divides the mask according to the pixels of each floating object, counts the number of pixels covered by each mask, converts the number of pixels into the actual physical area according to the camera calibration parameters, obtains the coverage area of ​​each floating object, and accumulates the coverage areas of each floating object to obtain the total coverage area of ​​the floating objects.

[0183] For example, the above process satisfies the following relationship:

[0184] in, This represents the physical area corresponding to the pixel segmentation mask for floating objects on the water surface, i.e., the total coverage area. Indicates the first The number of pixels contained in a pixel segmentation mask. This represents the actual physical area corresponding to a single pixel. This indicates the number of pixel segmentation masks, which is also the number of floating objects.

[0185] It is understandable that the actual coverage area of ​​a floating object is obtained by multiplying the number of pixels contained in the pixel segmentation mask of each floating object by the actual physical area of ​​a single pixel; the total coverage area of ​​all floating objects is obtained by summing up the actual coverage areas of all floating objects.

[0186] For example, assuming that after the vision sensor is calibrated, the actual physical size k corresponding to a single pixel is 0.01 meters / pixel, that is, the actual physical area corresponding to a single pixel is 0.0001 square meters.

[0187] At a certain moment, the visual recognition model detects three floating objects from the surface image. The number of pixels contained in the corresponding pixel segmentation masks are mask(1) = 5000 pixels, mask(2) = 8000 pixels, and mask(3) = 12000 pixels.

[0188] The actual coverage areas of each floating object are as follows: , , Therefore, the total coverage area is: 0.5 + 0.8 + 1.2 = 2.5 .

[0189] S1013. Determine the second feature based on the total coverage area, the total number of floating objects, and the area of ​​the water surface.

[0190] In this step, the computing device determines the coverage rate of the floating objects on the water surface based on the total coverage area of ​​the floating objects and the area of ​​the water surface; and determines the density of the floating objects on the water surface based on the total number of floating objects and the area of ​​the water surface.

[0191] In one possible implementation, the coverage rate is determined by the ratio of the total covered area to the surface area of ​​the water body.

[0192] For example, the process satisfies the following relationship:

[0193] in, C represents the total area of ​​the water body, and C represents the coverage rate.

[0194] In one possible implementation, density is determined as the ratio of the total number of floating objects to the surface area of ​​the water body. For example, this process satisfies the following relationship:

[0195] in, D represents the density of floating objects on the water surface identified after visual recognition processing. After standardization processing as shown in equation (3), we obtain D.

[0196] In this way, by processing surface images through a visual recognition model, the total coverage area and total number of floating objects can be determined, thereby determining the coverage rate and density, achieving efficient extraction and accurate quantification of the visual features of floating objects.

[0197] In some embodiments, the computing device can deploy the visual recognition model locally, wherein the training process is divided into two stages: first, a base model of the visual recognition model is used to pre-train on a general public dataset to enable the model to acquire general object detection and segmentation capabilities; then, the pre-trained model is fine-tuned on a floating object dataset to adapt to the specific recognition scenario of floating objects.

[0198] The general public dataset contains images of various categories of common objects, covering annotation information for object detection and pixel-level segmentation. The floating debris dataset contains images of floating debris in front of debris barriers. For example, sample images in the dataset may include water surface images under different lighting conditions, water surface images with different floating debris densities, and water surface images of different types of floating debris (such as branches, aquatic plants, and household waste). Each sample image is labeled with the category label of the floating debris, the location of the detection box, and a pixel-level segmentation mask. By training on these datasets, the visual recognition model can learn general feature representations of floating debris, thereby achieving accurate detection and segmentation of floating debris in practical applications.

[0199] The following describes the processing of surface images by the improved visual recognition model in the embodiments of this application.

[0200] In some embodiments, as described in S1012 above, the basic YOLOv8-Seg model includes a backbone network, a neck network, and a head network. In complex water surface scenarios, the basic model is susceptible to background interference, suffers from insufficient multi-scale feature fusion, and exhibits inadequate target edge segmentation accuracy, making it difficult to meet the requirements for accurate identification of floating objects on the water surface. Therefore, computing devices have improved the YOLOv8-Seg model. In exemplary embodiments, such as... Figure 7 As shown, the architecture of the improved visual recognition model includes: The backbone network includes a convolution module (Conv), a multi-dimensional attention convolution module (triplet-Conv), a cross-stage partial with feature fusion module (C2f), and a multi-scale pooling fusion module.

[0201] Among them, the multi-scale pooling fusion module can also be called the Fusion Fast Spatial Pyramid Pooling Module (FusionSPPF).

[0202] The backbone network has a serially stacked structure. The visual recognition model inputs the surface image into the backbone network. The surface image first undergoes preliminary feature extraction through a convolutional module to obtain an initial feature map. The initial feature map then alternately passes through multiple triplet-Conv modules and C2f modules for multi-level feature extraction. During the feature extraction process, attention weighting is performed through triplet-Conv, outputting a multi-level weighted feature map.

[0203] The feature map output from the deepest C2f module is input into the FusionSPPF module. The FusionSPPF module performs adaptive pooling and convolutional fusion on this deepest feature map to generate enhanced deep features, which then replace the original deepest feature map, forming an updated multi-layer weighted feature map. The updated multi-layer weighted feature maps are then input into the neck network via the C2f module.

[0204] The neck network consists of an upsampling module, a concat module, and a C2f module. The neck network adopts a bidirectional cross-scale fusion topology that combines top-down and bottom-up approaches. It receives multi-layer weighted feature maps output from the backbone network, performs feature fusion processing step by step, and finally outputs three fused feature maps of different scales in parallel to the head network.

[0205] The neck network receives the updated multi-layer weighted feature map output by the backbone network. First, it transmits deep semantic features to shallow layers through upsampling, feature concatenation, and the C2f module. Then, it aggregates shallow position features to deep layers through feature concatenation and the C2f module, completing bidirectional cross-scale feature fusion. This results in three independent multi-path fused feature maps with different scales, which are then output to the head network.

[0206] The head network includes a detection branch and a segmentation branch. The segmentation branch contains three parallel multi-branch reparameterization (DBB) modules; for ease of description, this segmentation branch and the DBB module are collectively referred to as segment-DBB. An NMS module is also configured. The three DBB modules operate independently and in parallel, each receiving a fused feature map output from the neck network.

[0207] The multi-path fused feature maps are fed into the detection branch and segment-DBB respectively. In the detection branch, the multi-path fused feature maps are directly used for target detection, outputting a set of candidate detection boxes. In segment-DBB, the segment-DBB module performs feature enhancement processing on each fused feature map; each branch performs pixel-level segmentation based on the enhanced feature map, outputting a set of candidate pixel segmentation masks. The candidate detection boxes and candidate pixel segmentation masks are aggregated to the NMS module, which removes duplicate candidate targets corresponding to the same floating object, finally outputting the detection boxes and pixel segmentation masks corresponding to each floating object in the surface image.

[0208] In an exemplary embodiment, such as Figure 8 As shown, triplet-Conv includes a 3D attention unit, a 2D batch normalization unit (BatchNorm2d), and a SiLu activation unit. The 3D attention unit adjusts the feature weights of the input feature map in both channel and spatial dimensions to enhance important features and suppress irrelevant features. Its output is normalized by the 2D batch normalization unit, and then a nonlinear transformation is introduced through the SiLu activation unit, ultimately outputting a weighted feature map.

[0209] Thus, the triplet-Conv module enables the network to adaptively focus on important features related to floating objects during the feature extraction stage, while suppressing background interference such as water surface reflection and shadows, thereby improving the model's feature extraction capability under complex lighting conditions.

[0210] In an exemplary embodiment, such as Figure 9 As shown, the processing procedure of the FusionSPPF module includes: The deepest weighted feature map output from the backbone network is received, and adaptive pooling is performed on the feature map at different scales. The feature maps obtained at different pooling scales are then input into the corresponding Covn module and LeakyReLU activation function for feature extraction and nonlinear mapping to enhance the expressive power of the features. Figure 9 (The feature maps at three scales are processed in the middle).

[0211] Next, the multi-scale feature maps obtained from the processing are concatenated to obtain the fused enhanced deep feature map. Finally, the fused feature map is processed sequentially through convolutional layers and max-pooling layers to generate the final enhanced deep feature map. This enhanced deep feature map is then used to replace the original deepest feature map in the multi-layer weighted feature map output by the backbone, resulting in the updated multi-layer weighted feature map.

[0212] Thus, the FusionSPPF module can enhance the deepest features of the backbone with multi-scale information, enrich the semantic features of large floating objects, and work with other scale features to improve the model's ability to perceive multi-scale targets, thereby improving the comprehensiveness and accuracy of detection.

[0213] In an exemplary embodiment, such as Figure 10 As shown, the processing procedure of the DBB module includes: The DBB module inputs one fused feature map from the neck network output into four parallel branches for processing.

[0214] The first branch (counting from left to right) involves the input feature map being convolved by a 1×1 convolutional layer (Conv) and then connected to a batch normalization (BN) layer for further processing, outputting the first feature.

[0215] The second branch processes the input feature map through a 1×1 Conv layer, followed by a BN layer, then through a K×K Conv layer, and finally through another BN layer to output the second feature map.

[0216] Where K represents the size of the convolution kernel.

[0217] The third branch processes the input feature map through a 1×1 Conv layer, followed by a BN layer, then an average pooling (AVG) layer, and finally another BN layer to output the third feature.

[0218] The fourth branch processes the input feature map through a K×K Conv layer and connects it to a BN layer to output the fourth feature.

[0219] The first, second, third, and fourth features are added together (ADD) to obtain a fused feature map, which serves as the enhanced feature map output by the DBB module. This enhanced feature map is then fed into the detection and segmentation branches for subsequent operations to generate floating object detection boxes and pixel segmentation masks. During the inference phase, the parameters of the four branches are equivalently converted to the parameters of a single-path K×K convolutional layer to reduce computational load and improve inference speed.

[0220] Thus, through the DBB module, richer details and texture features of floating object edges can be extracted during the training phase through a multi-branch structure, improving segmentation accuracy. At the same time, during the inference phase, it is reparameterized into a single-branch structure, ensuring inference efficiency and achieving a balance between accuracy and speed.

[0221] Based on the above description, this application implements a process for processing surface images based on an improved visual recognition model, such as... Figure 11 As shown, S1011 specifically includes the following steps S301-S305: S301. Through the multidimensional attention convolution module of the backbone network, feature extraction and attention weighting are performed on the surface image to obtain a multi-layer weighted feature map.

[0222] Attention-weighted processing is used to adjust the weights of the feature map along the channel and spatial dimensions during feature extraction to enhance the feature response corresponding to the floating object. The spatial dimensions include the spatial width and spatial height dimensions.

[0223] Feature extraction refers to the process of performing convolution operations by sliding convolution kernels across an image to extract feature information at different levels, such as edges, textures, and shapes. Correspondingly, multi-layer weighted feature maps refer to multiple feature maps output by convolutional layers of different depths in the backbone network. Each feature map corresponds to a different spatial resolution and semantic abstraction level. Shallow feature maps have higher spatial resolution and are suitable for capturing detailed information such as the edges and textures of floating objects, while deep feature maps have a larger receptive field and stronger semantic information, and are suitable for capturing the overall outline and category information of floating objects.

[0224] S302. Adaptive pooling and convolution fusion are performed on the deepest feature map in the multi-layer weighted feature map through the multi-scale pooling fusion module of the backbone network to update the multi-layer weighted feature map.

[0225] The deepest feature map represents the feature map with the lowest spatial resolution and the strongest semantic information in the backbone network. This feature map has the largest receptive field and can capture the overall semantic information and context of the floating object.

[0226] Adaptive pooling refers to dynamically adjusting the size and stride of the pooling window according to a preset output size, so that input feature maps of different sizes can be pooled to the same spatial size, thus preserving feature information at different scales.

[0227] Convolutional fusion refers to inputting multi-scale feature maps, after adaptive pooling, into convolutional layers for feature extraction, then concatenating and fusing the extracted multi-scale feature maps to obtain enhanced features. These enhanced features are then used to replace the deepest feature map in the original multi-layer weighted feature map, thus updating the multi-layer weighted feature map.

[0228] S303. The updated multi-layer weighted feature map is fused through the neck network to obtain a multi-path fused feature map.

[0229] Among them, fusion refers to the adoption of a bidirectional cross-scale fusion structure, which uses upsampling to transfer semantic information from deep feature maps to shallow feature maps, and uses downsampling to transfer spatial location information from shallow feature maps to deep feature maps, thereby achieving information complementarity and enhancement between feature maps of different scales.

[0230] Multi-path fusion feature maps refer to multiple enhanced feature maps of different scales output by the neck network, including shallow high-resolution feature maps, mid-level feature maps, and deep low-resolution feature maps, which are suitable for detecting floating objects of different sizes.

[0231] S304. The multi-branch reparameterization module of the head network performs feature enhancement processing on the multi-path fused feature map, and performs target detection and pixel segmentation processing based on the enhanced fused feature map to obtain the detection box and pixel segmentation mask corresponding to each floating object in the surface image.

[0232] Feature enhancement processing refers to extracting features from the input feature map under different receptive fields and different transform domains through four parallel branches of the multi-branch reparameterization module (including a 1×1 convolution branch, a 1×1 convolution and K×K convolution combination branch, a 1×1 convolution and average pooling combination branch, and a K×K convolution branch), and then adding and fusing the features output by each branch to enrich the expressive power and richness of the features.

[0233] Object detection refers to classifying objects and regressing bounding boxes on the enhanced fused feature map through detection branches to determine the location and category of floating objects.

[0234] Pixel segmentation refers to classifying pixels one by one on the enhanced fused feature map by segmenting branches, assigning a category label to each pixel, thereby generating the pixel-level outline of the floating object.

[0235] A pixel segmentation mask is a binary mask matrix corresponding to the size of the input image. The value of each pixel in the mask indicates whether the pixel belongs to a floating object. A value of 1 indicates that the pixel belongs to a floating object, and a value of 0 indicates that the pixel belongs to the background.

[0236] S305. Based on the detection boxes corresponding to each floating object, determine the total number of floating objects, and based on the pixel segmentation mask corresponding to each floating object, determine the total coverage area of ​​the floating objects.

[0237] The number of detection boxes represents the total number of floating objects. The method for determining the total coverage area can be found in the relevant description in S1012. That is, based on the number of pixels covered by each pixel segmentation mask and the mapping relationship between pixels and actual physical areas, the actual coverage area of ​​each floating object is determined. The coverage areas of all floating objects are then summed to obtain the total coverage area of ​​the floating objects.

[0238] Thus, by introducing a multi-dimensional attention convolution module into the backbone network, a multi-scale pooling fusion module between the backbone network and the neck network, and a multi-branch reparameterization module into the head network, the model's feature extraction capability, multi-scale perception capability, and pixel-level segmentation accuracy under complex lighting conditions can be enhanced, thereby improving the detection and segmentation performance of the visual recognition model in complex water surface environments.

[0239] The following section describes the process for determining the aggregation coefficient of floating objects.

[0240] In some embodiments, in S104, the computing device can also dynamically adjust the weight values ​​corresponding to the flow blockage coefficient and the surface density coefficient according to the current environmental conditions of the water body. For example... Figure 12 As shown, S104 may specifically include the following steps S1041-S1042: S1041. Based on preset rules, determine the first weight value of the overcurrent blockage coefficient and the second weight value of the surface density coefficient.

[0241] The sum of the first weight value and the second weight value is 1.

[0242] In some embodiments, the preset rule can be to dynamically adjust the relationship between the first weight value and the second weight value based on the image quality of the water surface image. Since the flow obstruction coefficient originates from hydraulic sensor data and the surface density coefficient originates from the visual recognition model's processing result of the surface image, their reliability differs under different environmental conditions. When the surface image quality is high, the reliability of the visual recognition result is high; in this case, the second weight value corresponding to the surface density coefficient should be greater than the first weight value corresponding to the flow obstruction coefficient. Conversely, when the surface image quality is low, the reliability of the visual recognition result decreases; in this case, the first weight value corresponding to the flow obstruction coefficient should be greater than the second weight value corresponding to the surface density coefficient.

[0243] In this case, the preset rules include: if the image quality of the surface image of the water body meets the quality conditions, the first weight value is greater than the second weight value; otherwise, the first weight value is less than the second weight value.

[0244] For example, the quality conditions may include at least one of the following: image sharpness greater than a preset sharpness threshold, image contrast greater than a preset contrast threshold, or image signal-to-noise ratio greater than a preset signal-to-noise ratio threshold.

[0245] For example, if the image quality of the surface image of the water body meets the quality requirements, the computing device can set a first weight value of 0.6 and a second weight value of 0.4. Otherwise, the first weight value can be set to 0.4 and the second weight value to 0.6.

[0246] In one possible implementation, when the visual sensor experiences hardware failure, image acquisition failure, or abnormal inference in the visual recognition model, the surface density coefficient becomes unreliable. In this case, the second weight value is set to 0, and the first weight value is set to 1, relying on hydraulic sensor data such as liquid level and flow rate for judgment. Conversely, when the liquid level sensor or flow sensor experiences hardware failure, abnormal data acquisition, or communication interruption, the overflow blockage coefficient becomes unreliable. In this case, the first weight value is set to 0, and the second weight value is set to 1, relying on the visual recognition result for judgment.

[0247] In other embodiments, the computing device can also directly acquire the ambient light intensity of the water body. In this case, the preset rules include: when the ambient light intensity is greater than a preset light threshold, it indicates that the water surface image is less affected by light, and the reliability of the visual recognition result is high; in this case, the first weight value is less than the second weight value. When the ambient light intensity is less than or equal to the preset light threshold, it indicates that the water surface image is affected by insufficient light, and the reliability of the visual recognition result decreases; in this case, the first weight value is greater than the second weight value. The light intensity can be acquired by a photosensitive sensor deployed on-site in the water body, or brightness information can be extracted from the image acquired by the visual sensor as a representation value of the light intensity. This application does not limit the method of acquiring the light intensity.

[0248] S1042. Based on the first weight value and the second weight value, the flow blockage coefficient and the surface density coefficient are weighted and summed to determine the floating matter aggregation coefficient of the water body.

[0249] For example, the process satisfies the following relationship:

[0250] in, This represents the first weight value. This represents the second weight value.

[0251] For example, assuming the image quality of the current water surface image meets the quality requirements, the computing device determines a first weight value w1 = 0.6 and a second weight value w2 = 0.4 according to preset rules. If the flow blockage coefficient S = 0.35 and the surface density coefficient V = 0.82, then the floating debris aggregation coefficient G = 0.6 × 0.35 + 0.4 × 0.82 = 0.21 + 0.328 = 0.538.

[0252] For example, assuming the image quality of the current water surface image does not meet the quality requirements, the computing device determines a first weight value w1 = 0.4 and a second weight value w2 = 0.6 according to preset rules. If the flow blockage coefficient S = 0.35 and the surface density coefficient V = 0.82, then the floating matter aggregation coefficient G = 0.4 × 0.35 + 0.6 × 0.82 = 0.14 + 0.492 = 0.632.

[0253] For example, suppose the visual recognition model malfunctions, making the surface density coefficient unreliable. The computing device determines the first weight value w1 = 1.0 and the second weight value w2 = 0. If the overflow blockage coefficient S = 0.45, then the floating object aggregation coefficient G = 1.0 × 0.45 + 0 × V = 0.45.

[0254] For example, assuming a malfunction in the level sensor renders the overflow blockage coefficient unreliable, the calculation device determines the first weight value w1 = 0 and the second weight value w2 = 1.0. If the surface density coefficient V = 0.78, then the floating matter aggregation coefficient G = 0 × S + 1.0 × 0.78 = 0.78.

[0255] Thus, by dynamically adjusting the weights according to image quality during the fusion process and resetting the corresponding weights to zero when the sensor or vision model fails, the fusion result can adapt to different environmental conditions, while maintaining basic control capabilities even when a single mode fails, thereby improving the robustness and reliability of control decisions.

[0256] In the embodiments of this application, for comparisons between each parameter and a preset threshold, "equal to" can be attached to "greater than" or "less than". For example, "greater than or equal to" and "greater than" can be configured to have the same triggering effect in a specific implementation, and the same applies to "less than or equal to" and "less than". The embodiments of this application do not limit the specific value selection method of the comparison operator.

[0257] Figure 13 This is a schematic diagram of a cleaning equipment control device provided in an embodiment of this application. Figure 13 As shown, the device includes: The acquisition module 1301 acquires a first feature of the water body and a second feature of floating objects on the water surface. The first feature includes the water level and flow rate at the trash rack, and the second feature includes the coverage of the floating objects on the water body and the density of the floating objects.

[0258] Processing module 1302 determines the flow obstruction coefficient of the water body based on a first feature. The flow obstruction coefficient characterizes the degree of obstruction caused by floating debris accumulating at the trash rack, thus hindering the trash rack's flow capacity. Based on a second feature, it determines the surface density coefficient of the floating debris. The surface density coefficient characterizes the density of floating debris on the water body surface. Based on the flow obstruction coefficient and the surface density coefficient, it determines the floating debris aggregation coefficient of the water body. If the floating debris aggregation coefficient exceeds a preset first aggregation coefficient threshold, it controls the cleaning equipment to start.

[0259] Other optional implementation methods are described in the corresponding sections of the above method embodiments, and will not be repeated here.

[0260] This application also provides a computing device. For example... Figure 14As shown, the computing device includes: a memory 1401, a transceiver 1402, and at least one processor 1403.

[0261] Transceiver 1402 is used to acquire flow data, liquid level data, and surface image data.

[0262] The memory 1401 stores computer program code, which includes computer instructions. These computer instructions run in the aforementioned computing device to implement the method shown in the above-described method embodiments.

[0263] For example, the memory may include high-speed random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.

[0264] Processor 1403 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Processor 1403 can also be other general-purpose processors. The general-purpose processor can be a microprocessor or any conventional processor.

[0265] The memory 1401, transceiver 1402, and processor 1403 are communicatively connected. For example, the memory 1401 and transceiver 1402 can be connected to the processor 1403 via a system bus and communicate with each other. The system bus can be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, an industry standard architecture (ISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 14 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0266] Alternatively, the memory 1401 can be either standalone or integrated with the processor 1403. When the memory 1401 is set up independently, it is connected to the processor 1403 via a system bus.

[0267] This application also provides a computer-readable storage medium. All or part of the processes in the above method embodiments can be executed by computer instructions instructing related hardware. The program can be stored in the aforementioned computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. The computer-readable storage medium can be any of the foregoing embodiments or memory. The aforementioned computer-readable storage medium can also be an external storage device of the recovery device, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the recovery device. Further, the aforementioned computer-readable storage medium can include both internal storage units of the recovery device and external storage devices. The aforementioned computer-readable storage medium is used to store the aforementioned computer program and other programs and data required by the recovery device. The aforementioned computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0268] This application also provides a computer program product comprising a computer program that, when run on a computer, causes the computer to perform any of the methods provided in the above embodiments.

[0269] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0270] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

[0271] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for controlling a cleaning and decontamination device, characterized in that, The method includes: Acquire a first characteristic of the water body and a second characteristic of the floating objects on the surface of the water body; the first characteristic includes the water level and flow rate at the trash rack, and the second characteristic includes the coverage of the floating objects on the water body and the density of the floating objects. Based on the first feature, the flow blockage coefficient of the water body is determined; the flow blockage coefficient characterizes the degree of blockage caused by the accumulation of floating objects at the trash rack, affecting the flow capacity of the trash rack. Based on the second feature, the surface density coefficient of the floating object is determined; the surface density coefficient characterizes the density of the floating object on the water surface. The floating debris aggregation coefficient of the water body is determined based on the flow blockage coefficient and the surface density coefficient. If the floating debris aggregation coefficient is greater than a preset first aggregation coefficient threshold, the cleaning equipment is activated.

2. The method according to claim 1, characterized in that, The step of controlling the cleaning equipment to start when the floating debris aggregation coefficient is greater than a preset first aggregation coefficient threshold includes: If the first condition is met and the floating object aggregation coefficient is greater than a preset first aggregation coefficient threshold, the cleaning equipment is controlled to start. The first condition includes any one of the following: The coverage rate is greater than a preset first coverage threshold; The density is greater than a preset first density threshold; The liquid level difference at the trash rack is greater than a preset first liquid level difference threshold; or; The flow rate at the trash rack is greater than a preset first flow rate attenuation threshold and the density is greater than a preset second density threshold.

3. The method according to claim 1, characterized in that, The method further includes: During the operation of the cleaning equipment, the floating debris aggregation coefficient is re-determined; If the redefined floating debris aggregation coefficient is less than the second aggregation coefficient threshold and a preset second condition is met, the cleaning equipment is controlled to stop; the second aggregation coefficient threshold is less than the first aggregation coefficient threshold. The second condition includes: the coverage rate is less than a preset second coverage rate threshold, the density is less than a third density threshold, the liquid level difference at the trash rack is less than a preset second liquid level difference threshold, and the flow rate attenuation at the trash rack is less than a preset second flow rate attenuation threshold.

4. The method according to any one of claims 1-3, characterized in that, Obtaining a second characteristic of the floating objects on the water surface includes: Obtain a surface image of the water body; The surface image is processed using a visual recognition model to identify the total coverage area of ​​the floating objects and the total number of the floating objects; and The second feature is determined based on the total coverage area, the total number of floating objects, and the area of ​​the water surface.

5. The method according to claim 4, characterized in that, The visual recognition model includes a backbone network, a neck network, and a head network; wherein the backbone network includes a multi-dimensional attention convolution module and a multi-scale pooling fusion module, and the head network includes a multi-branch reparameterization module. The step of processing the surface image using a visual recognition model to identify the total coverage area of ​​the floating objects and the total number of the floating objects includes: The surface image is processed by the multidimensional attention convolution module of the backbone network to extract features and perform attention weighting, resulting in a multi-layer weighted feature map. The attention weighting process is used to adjust the weights of the feature map along the channel dimension and spatial dimension during the feature extraction process to enhance the feature response corresponding to the floating object. The deepest feature map in the multi-layer weighted feature map is adaptively pooled and convolutionally fused using the multi-scale pooling fusion module of the backbone network to update the multi-layer weighted feature map. The updated multi-layer weighted feature map is fused using the neck network to obtain a multi-path fused feature map; The multi-branch reparameterization module of the head network performs feature enhancement processing on the multi-path fused feature map, and performs target detection and pixel segmentation processing based on the enhanced fused feature map to obtain the detection box and pixel segmentation mask corresponding to each floating object in the surface image. Based on the detection bounding boxes corresponding to each of the floating objects, the total number of the floating objects is determined, and based on the pixel segmentation mask corresponding to each of the floating objects, the total coverage area of ​​the floating objects is determined.

6. The method according to any one of claims 1-3, characterized in that, The process of fusing the flow blockage coefficient and the surface density coefficient to determine the floating matter aggregation coefficient of the water body includes: Based on preset rules, a first weight value for the overcurrent blockage coefficient and a second weight value for the surface density coefficient are determined; the sum of the first weight value and the second weight value is 1. The flow blockage coefficient and the surface density coefficient are weighted and summed based on the first weight value and the second weight value to determine the floating debris aggregation coefficient of the water body. The preset rules include: if the image quality of the surface image of the water body meets the quality conditions, the first weight value is greater than the second weight value; otherwise, the first weight value is less than the second weight value.

7. A control device for a cleaning and decontamination equipment, characterized in that, The device includes: The acquisition module acquires a first feature of the water body and a second feature of the floating objects on the surface of the water body; the first feature includes the water level and flow rate at the trash rack, and the second feature includes the coverage of the floating objects on the water body and the density of the floating objects. The processing module determines the flow obstruction coefficient of the water body based on the first feature; the flow obstruction coefficient characterizes the degree of obstruction caused by the accumulation of floating objects at the trash rack, affecting the flow capacity of the trash rack; it determines the surface density coefficient of the floating objects based on the second feature; the surface density coefficient characterizes the density of the floating objects on the surface of the water body; it determines the floating object aggregation coefficient of the water body based on the flow obstruction coefficient and the surface density coefficient; and it controls the cleaning equipment to start when the floating object aggregation coefficient is greater than a preset first aggregation coefficient threshold.

8. A computing device, characterized in that, include: A memory and at least one processor; the memory is communicatively connected to the processor; the memory is used to store computer program code, the computer program code including computer instructions; when the processor executes the computer instructions, the computing device performs the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, are used to implement the method as described in any one of claims 1-6.

10. A computer program product, characterized in that, When the computer program product is run on a computer / executed by the computer's processor, it implements the method as described in any one of claims 1-6.

Images