Intelligent monitoring device and method for water environment

Abstract

The invention discloses an intelligent monitoring device and method for a water body environment. The monitoring device comprises a platform base body (5), a propeller assembly (7), a water quality detector (1), a camera (4), a detection box (9), a main winch (32) and a main controller (31). The detection box is assembled and connected with a hoisting steel rope (321) of the main winch, and the main winch can lower and hoist the detection box; the propeller assembly is used for driving the monitoring device to autonomously move on the water surface. According to the monitoring device and method, automatic monitoring can be flexibly and efficiently carried out on the water body of the water area, aiming at a method for monitoring the floating objects on the water surface through machine vision, a two-dimensional wavelet convolution (WTConv2d) technology is used, multiple attention mechanisms are added, so that better feature representation is obtained, 5% of parameter quantity and calculation quantity of the model are reduced, and the accuracy of the model is improved. The calculation complexity of a water surface target detection task is reduced; and the model understanding capability is improved.

Description

Technical Field

[0001] The present invention relates to a water quality detection technology, and particularly to an intelligent monitoring device and method for water environment. Background Art

[0002] With the acceleration of the industrialization and urbanization processes, various pollutants are discharged into water bodies in large quantities. Water quality monitoring devices have important application values in the fields of water area environmental protection, aquaculture, agricultural irrigation, etc. Existing water quality monitoring devices have problems such as high maintenance costs, poor flexibility, inaccurate monitoring data, and unstable data transmission. The main manifestations are as follows: The water quality sensors are placed in water for a long time, and algae, shellfish, microorganisms, etc. are easily attached to their surfaces, and regular cleaning is required to ensure the accuracy of monitoring data; Fixed single-body buoys can only monitor at fixed points and cannot reflect water quality indicators in a large range; There are differences in water quality parameters in the depth direction of water, and fixed sensors cannot obtain stratified water quality parameters; Contact water quality sensors have deficiencies in detecting parameters such as the concentration of suspended solids in water, heavy metal ions, organic matter content, and nutrient salt content; A single communication signal will cause problems such as inaccurate data and data loss.

[0003] Regarding environmental monitoring on the water surface. Traditional methods for monitoring floating objects on the water surface, such as manual visual inspections and conventional satellite remote sensing monitoring, have obvious limitations. Manual inspections are inefficient and have a limited coverage area, making it difficult to achieve real-time and comprehensive monitoring; Although satellite remote sensing monitoring has a wide coverage area, its resolution is limited, it is difficult to accurately identify small floating objects, and it is greatly affected by factors such as weather. In the context of the rapid development of computer vision technology, object detection algorithms have become a powerful tool for solving such problems. The YOLO series of algorithms have been widely used in many object detection scenarios due to their advantages such as fast detection speed and relatively high accuracy. However, the original YOLO model has problems such as a high computational complexity and insufficient detection accuracy for small targets and partially occluded targets when facing complex water surface environments.

[0004] In summary, the current problems are as follows:

[0005] 1) The water quality sensors are placed in water for a long time, and algae, shellfish, microorganisms, etc. are easily attached to their surfaces, and regular cleaning is required to ensure the accuracy of monitoring data;

[0006] 2) Fixed single-body buoys can only monitor at fixed points and cannot reflect water quality indicators in a large range;

[0007] 3) Fixed sensors cannot obtain stratified water quality parameters;

[0008] 4) Contact water quality sensors have deficiencies in detecting parameters such as the concentration of suspended solids in water, heavy metal ions, organic matter content, and nutrient salt content;

[0009] 5) A single communication signal will cause problems such as inaccurate data and data loss.

[0010] 6) When the original YOLO model faces the complex water surface environment, there are problems such as high computational complexity and insufficient detection accuracy for small targets and partially occluded targets. Summary of the Invention

[0011] The purpose of the present invention is to provide an intelligent monitoring device and method for water environment, which can automatically and efficiently monitor water bodies in waters in a flexible and mobile manner.

[0012] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0013] An intelligent monitoring device for water environment, the monitoring device includes a platform base body, a thruster assembly, a water quality detector, a camera, a detection box, a main winch and a main controller; the thruster assembly, the water quality detector, the camera, the main winch and the main controller are all installed based on the platform base body; the detection box is assembled and connected with the winch cable of the main winch, and the main winch can lower and wind the detection box; the thruster assembly is used to drive the monitoring device to move autonomously on the water surface.

[0014] Further, a water quality sensor, an auxiliary controller and an auxiliary lithium battery are arranged in the detection box.

[0015] Further, both the main controller and the auxiliary controller are configured with a wireless communication module, and the main controller and the auxiliary controller can perform wireless data communication. The main controller can send control instructions to the auxiliary controller wirelessly, and the auxiliary controller can transmit the detection data of the water quality sensor to the main controller wirelessly.

[0016] Further, when the detection box is in the water body, the communication method between the main controller and the auxiliary controller adopts underwater acoustic communication; when the detection box leaves the water body, the communication method between the main controller and the auxiliary controller adopts Lora communication.

[0017] Further, a detection body berthing part is provided on the platform base body, and the detection body berthing part is used to berth the detection box.

[0018] Further, a main electromagnet, a wireless charging transmitting coil and a limit switch are arranged at the detection body berthing part, and a passive magnet and a wireless charging receiving coil are arranged on the detection box; when the detection box berths at the detection body berthing part, the limit switch is triggered, the main electromagnet is energized to generate magnetic force, the main electromagnet and the passive magnet attract each other, the wireless charging transmitting coil and the wireless charging receiving coil are correspondingly close together, and the detection box can be positioned and berthed at the detection body berthing part and realize wireless charging.

[0019] Furthermore, a ranging radar is provided at the bottom of the detection box; the monitoring device is configured with a gyroscope and a GPS positioning component.

[0020] An intelligent monitoring method for water environment, which uses the monitoring device as described above; the monitoring method uses a water surface floating object recognition model to recognize water surface floating objects; the water surface floating object recognition model is constructed based on an improved YOLOv11 model; the water surface floating object recognition model includes a feature extraction network Backbone module, a feature fusion network Neck module, and a feature recognition network Head module; the C3K2 module in the feature extraction network Backbone module is replaced with a C3K2_WTConv module that integrates a two-dimensional wavelet convolution algorithm; in the C3K2_WTConv module, the original Bottleneck module is replaced with a Bottleneck_WTConv module; a two-dimensional wavelet convolution feature processing algorithm option is added in the Bottleneck_WTConv module, and whether to use the two-dimensional wavelet convolution algorithm is selected according to the intermediate channel number after dimensionality reduction of the input channel number; the feature fusion network Neck module embeds a triple attention mechanism module Triplet Attention in the small target detection layer.

[0021] Furthermore, the training process of the water surface floating object recognition model includes: constructing a water surface floating object image data set, which includes a training set, a validation set, and a test set; all target objects in the water surface floating object image data set are labeled with corresponding bounding boxes, and corresponding labels for indicating the categories of water surface floating object images are set; the water surface floating object training set images are respectively input into a network model improved based on YOLOv11 for training to obtain a trained water surface floating object recognition model; the trained water surface floating object recognition model is verified using the water surface floating object validation set images; the trained water surface floating object recognition model is evaluated for performance using the water surface floating object test set, and the performance evaluation includes calculating accuracy, recall, and F1 score metrics.

[0022] Furthermore, in the two-dimensional wavelet convolution feature processing algorithm, wavelet transform is combined with convolution, integrating the advantages of wavelet transform in capturing features of different frequencies and traditional convolution in extracting spatial features.

[0023] The beneficial effects of the monitoring device of the present invention compared with the prior art are as follows:

[0024] 1) The mobility and flexibility are improved. The unique underwater thruster layout enables omnidirectional flexible movement and stable operation, expanding the monitoring range;

[0025] 2) Reduce costs and facilitate maintenance. The detection box is separated from the platform base and transmits data wirelessly, reducing cable damage and maintenance costs;

[0026] 3) Achieve efficient wireless charging. The structural setting of the main electromagnet and the passive magnet solves the problem of alignment between the detection box and the upper platform coil, improves charging efficiency and safety, and ensures stable energy supply for the detection box;

[0027] 4) Accurate recovery positioning. The setting of the limit switch helps to accurately judge whether the detection box has reached the berthing part of the detection body during recovery and control the main winch to stop the winch, thereby improving the recovery accuracy and reliability and reducing equipment damage and uncertainty of manual intervention;

[0028] 5) Realize automatic and precise control, set up a ranging radar, so that the detection box can automatically and accurately sense the distance to obstacles and adjust the servo motor when it is lowered, improve safety and accuracy, and reduce the risk of collision;

[0029] 6) Enhance the reliability of unmanned operations. The automated design of lowering and recovering reduces reliance on manpower, reduces the probability of human errors and failures, facilitates stable operation in complex underwater environments, expands the scope of application and adaptability, and provides strong support for underwater unmanned detection missions;

[0030] 7) The contact water quality sensor analyzes the water quality at different depths layer by layer, and in order to ensure the stability and multi-dimensionality of the data, it integrates the data of the non-contact water quality detector to provide more comprehensive and accurate water quality information;

[0031] 8) Adopt multiple communication methods, with strong adaptability and the ability to cope with complex environments; improve data transmission stability and complement shortcomings; meet diversified data transmission needs and adapt to different data; improve system reliability and flexibility, multi-link guarantee and flexible resource allocation;

[0032] 9) With the help of two-dimensional wavelet convolution (WTConv2d) technology, better feature representation is obtained, and the model reduces 5% of parameters and calculations, thereby reducing the computational complexity of the surface target detection task;

[0033] 10) Introducing the Triplet Attention mechanism in the small target detection layer enables the model to build the interdependence between channels and spatial positions, improving the model's ability to understand the characteristics of small targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a structural schematic diagram of a water environment intelligent monitoring device based on implementation mode 1 of the present invention;

[0035] Figure 2 This is a schematic diagram of the monitoring device of Embodiment 1 from an oblique downward angle, in which the housing of the main control compartment is removed;

[0036] Figure 3 Schematic diagram of the detection box in the monitoring device of Embodiment 1 being recovered and leaving the water body;

[0037] Figure 4 Schematic diagram at the upper platform of the monitoring device of Embodiment 1;

[0038] Figure 5 Exploded view of the detection box in the monitoring device of Embodiment 1;

[0039] Figure 6 Schematic structural diagram of the intelligent water environment monitoring device according to Embodiment 2 of the present invention;

[0040] Figure 7 Partial explosion view of the monitoring device of Embodiment 2;

[0041] Figure 8 Structural diagram of the detector recovery state of the monitoring device of Embodiment 2;

[0042] Figure 9 Internal structural diagram of the top - layer control box of the monitoring device of Embodiment 2;

[0043] Figure 10 Structural diagram of the connection between the detection box and the device platform of the monitoring device of Embodiment 2;

[0044] Figure 11 Structural diagram of the connection between the inside of the detection box and the device platform of the monitoring device of Embodiment 2;

[0045] Figure 12 Side view of the water quality monitoring device of the monitoring device of Embodiment 2;

[0046] Figure 13 Internal structural diagram of the underwater detection box of the monitoring device of Embodiment 2;

[0047] Figure 14 Attitude and position control flow chart of the monitoring device of the present invention;

[0048] Figure 15 Schematic diagram of wireless data transmission of the monitoring device of the present invention;

[0049] Figure 16 Schematic diagram of the monitoring process of surface floating objects of the monitoring device of the present invention;

[0050] Figure 17 Structural diagram of the improved model network of the monitoring device of the present invention;

[0051] Figure 18 Structural diagram of the C3K2_WTConv module of the monitoring device of the present invention;

[0052] Figure 19 Structural diagram of the Bottleneck_WTConv module of the monitoring device of the present invention;

[0053] Figure 20 Structural diagram of the WTConv2d module of the monitoring device of the present invention;

[0054] Figure 21 P-R curves of different networks of the monitoring device of the present invention,

[0055] wherein,

[0056] Sub-figure (1) is the detection effect diagram of water surface targets based on the improved YOLOv11 model,

[0057] Sub-figure (2) is the detection effect diagram of water surface targets of the traditional YOLOv11 model,

[0058] Sub-figure (3) is the detection effect diagram of water surface targets of the traditional YOLOv10 model,

[0059] Sub-figure (4) is the detection effect diagram of water surface targets of the traditional YOLOv8 model;

[0060] Figure 22 Schematic diagram of the actual detection result of water surface floating objects of the monitoring device of the present invention. Detailed implementation manners

[0061] The following further illustrates the present invention with specific embodiments:

[0062] Embodiment 1:

[0063] This Embodiment 1 provides an intelligent water environment monitoring device, which is used to monitor the water quality of a water area.

[0064] Referring to Figure 1 , the monitoring device of this Embodiment 1 mainly includes a platform base 5, a main control cabin 3, a thruster assembly 7, a photovoltaic panel 8, a water quality detector 1, a camera 4, an anchoring device 6, and a detection box 9.

[0065] The platform base 5 serves as the carrier of the entire monitoring device, and all device installations are based on the platform base 5, that is, the main control cabin 3, the thruster assembly 7, the photovoltaic panel 8, the water quality detector 1, the camera 4, the anchoring device 6, and the detection box 9 are all installed on the platform base 5.

[0066] More specifically,

[0067] The platform base 5 has a lower platform 51 and an upper platform 52. The platform base 5 also has a floating body 53, which is arranged below the lower platform 51 at the bottom part of the platform base 5. The floating body 53 is used to provide buoyancy for the platform base 5 and even the entire monitoring device, so as to ensure that the monitoring device can float on the water body.

[0068] The thruster assembly 7 and the anchoring device 6 are installed based on the lower platform 51.

[0069] There are two groups of the thruster assembly 7 in total. Both groups of the thruster assembly 7 are installed on the lower side of the lower platform 51. Each group of the thruster assembly 7 is composed of several thrusters. Its function is to drive the entire monitoring device to move autonomously on the water surface and maintain the water surface stability of the monitoring device.

[0070] In this embodiment, there are 8 thrusters in each group of the thruster assembly 7, including forward and backward thrusters, stabilizing thrusters, lateral thrusters, etc. The propulsion directions of the 8 thrusters are all different. Combining the attitude and position information fed back by the gyroscope and the GPS satellite navigation system, after the main controller performs PID operation, it controls the output of the thrusters to drive the monitoring device to move flexibly and stably on the water surface.

[0071] The anchoring device 6 is mainly composed of an anchor winch 61 and an anchor 62. The anchor winch 61 is installed on the lower platform 51. The anchor winch 61 is used to lower and wind the anchor 62 to achieve the functions of anchoring position and weighing anchor.

[0072] The camera 4 and the photovoltaic panel 8 are both installed on the upper platform 52. The photovoltaic panel 8 is used to provide additional power supply for the entire monitoring device, especially for charging the lithium battery. The camera 4 is used to provide video image monitoring of the water surface condition.

[0073] The main control cabin 3 is installed on the upper platform 52. The water quality detector 1 is installed based on the main control cabin 3. The water quality detector 1 used is a hyperspectral water quality detector, which belongs to non-contact water quality detection. The water quality detector 1 is used to detect and analyze parameters such as the concentration of suspended solids in the water body, heavy metal ions, organic matter content, and nutrient salt content by analyzing the spectral characteristics of reflected light or transmitted light.

[0074] See Figure 2 , a main controller 31 and a main winch 32 are arranged in the main control cabin 3, and a main lithium battery 33 is also equipped. The detection box 9 is assembled and connected to the winch cable 321 of the main winch 32. The main winch 32 is used to lower and wind the detection box 9 based on the platform base 5, so as to lower the detection box 9 into the water body, and then be able to lift the detection box 9 out of the water body, that is, recover the detection box 9 into the monitoring device.

[0075] Specifically,

[0076] A through hole is opened at the central part of the upper platform 52 as a cable passing hole. The diameter of the cable passing hole is smaller than the diameter of the detection box 9, so the detection box 9 cannot pass through the cable passing hole, and only the hoisting steel cable 321 for suspending the detection box 9 can pass through;

[0077] A through hole is opened at the central part of the lower platform 51 as a box passing hole. The diameter of the box passing hole is larger than the diameter of the detection box 9, so the detection box 9 can pass through the box passing hole, and the hoisting steel cable 321 for suspending the detection box 9 can also pass through;

[0078] The hoisting steel cable 321 of the main hoist 32 is lowered from the cable passing hole to the lower part of the upper platform 52, and then connected to the detection box 9. In this way, the main hoist 32 can lower and hoist the detection box 9 below the upper platform 52. The detection box 9 can be lowered (through the box passing hole) from the upper platform 52 into the water body (such as Figure 1 the state shown), and then can be hoisted and recovered from the water body to the lower part of the upper platform 52 (detached from the water body, such as Figure 3 the state shown).

[0079] It should be noted that, for the convenience of description, the contact part between the lower surface of the upper platform 52 and the hoisted detection box 9 is defined as the "detection body berthing part", such as Figure 4 the place indicated by the arrow A in

[0080] The detection box 9 is assembled and connected to the hoisting steel cable 321 of the main hoist 32, and the detection box 9 can be lifted to the detection body berthing part, and the surrounding is designed without obstruction, which is convenient for sunlight irradiation to relieve the adhesion of underwater organisms to the surface of the sensor.

[0081] The main controller 31 is used to control the overall operation of the entire monitoring device.

[0082] Specifically,

[0083] All electrical devices on the monitoring device, such as the water quality detector 1, the main hoist 32, the thruster assembly 7, the anchoring device 6, the camera 4, etc., are electrically connected to the main controller 31. All electrical devices on the monitoring device are controlled by or receive signals from the main controller 31, so as to realize the overall control of the entire monitoring device. For the detection box 9 mentioned later, the electrical devices therein are also controlled by or receive signals from the main controller 31. In addition, a main lithium battery 33 is equipped in the main control cabin 3, and the main lithium battery 33 is used to provide the main power supply for the entire monitoring device.

[0084] It should be noted that the photovoltaic panel 8 is connected to the main lithium battery 33. In the presence of sunlight, the photovoltaic panel 8 can provide sufficient charging power for the main lithium battery 33.

[0085] The monitoring device of this embodiment is also equipped with a gyroscope, which is arranged in the main control cabin 3. The gyroscope is used for: the main controller 31 can read the attitude angle of the device through the gyroscope.

[0086] The monitoring device of this embodiment is also equipped with a GPS positioning component, which is arranged in the main control cabin 3, so as to facilitate positioning the location of the monitoring device.

[0087] Combining the "attitude angle" and "positioning location" data, and combining with the expected attitude value of the operating state of the device, after performing PID operation, the parameters that need to be adjusted for the electric drive are obtained to control the output of the thruster assembly 7.

[0088] See Figure 5 , the detection box 9 is composed of a box body 91, a box bottom 92 and a box cover 93 assembled together.

[0089] A spiral buckle 931 is arranged on the box cover 93 to facilitate connection and assembly with the hoisting steel cable 321 of the main winch 32.

[0090] Inside the detection box 9 (inside the box body 91), there are many various types of water quality sensors 98, and an auxiliary controller 96 and an auxiliary lithium battery 97 are configured for these water quality sensors 98.

[0091] The many various types of water quality sensors 98 include water temperature sensors, dissolved oxygen sensors, pH value sensors, etc., which belong to contact type water quality detection. In this way, through these many water quality sensors 98, the detection box 9 can detect water quality indicators such as water temperature, dissolved oxygen, and pH value at different depths.

[0092] The contact type detection can conduct local monitoring of water bodies at different depths, and the non-contact type detection can quickly scan large areas of water. The contact type detection can accurately measure indicators such as water temperature, dissolved oxygen, and pH value, while the non-contact type detection can detect the concentration of suspended solids, heavy metal ions, organic matter content, and nutrient salt content in the water body. The combination of the two can obtain richer water quality parameters, comprehensively reflect the water body condition, and provide sufficient data support for comprehensively evaluating the water quality. In addition, the combination of the two methods can not only obtain stratified water quality data but also achieve large-area monitoring, improve the monitoring efficiency, and comprehensively master the water quality condition of the water area. When monitoring a river, the contact type detection can obtain water quality information at different depths through the liftable detection box 9, and the non-contact type detection can quickly obtain the water quality information of a large range on the river surface, realizing comprehensive and efficient monitoring of the river.

[0093] The water quality sensor 98 adopts a replaceable design so that different types or precisions of sensors can be replaced according to actual monitoring requirements, improving the adaptability of the monitoring system.

[0094] It should be noted that a large number of water permeable holes are provided on the box body 91, so that water can enter the detection box 9 through the water permeable holes and come into contact with the water quality sensor 98 therein.

[0095] A wireless charging receiving coil 94 is provided in the box cover 93, and the wireless charging receiving coil 94 is connected to the auxiliary lithium battery 97. Correspondingly, referring to Figure 4 , a wireless charging transmitting coil 522 is provided at the detection body berthing part of the upper platform 52, and the wireless charging transmitting coil 522 receives power supply from the photovoltaic panel 8. In this way, the photovoltaic panel 8 can charge the auxiliary lithium battery 97 through the wireless charging receiving coil 94 and the wireless charging transmitting coil 522.

[0096] A passive magnet 95 is provided in the box cover 93. Correspondingly, a main electromagnet 521 is provided at the detection body berthing part of the upper platform 52. When the box cover 93 of the detection box 9 abuts against the detection body berthing part of the upper platform 52, under the action of "the passive magnet 95 and the main electromagnet 521 attracting each other", the detection box 9 can be firmly adsorbed at the detection body berthing part of the upper platform 52. Since both the passive magnet 95 and the main electromagnet 521 are circular rings and have the same configuration size, they can achieve a good positioning effect when adsorbing. In this way, it is beneficial for the detection box 9 to be accurately positioned and berthed at the detection body berthing part, so that the wireless charging receiving coil 94 and the wireless charging transmitting coil 522 can be aligned with each other, thus achieving an effective and reliable wireless charging effect.

[0097] A limit switch 523 is also provided at the detection body berthing part of the upper platform 52, and the limit switch 523 is used to detect whether the detection box 9 is berthed at the detection body berthing part of the upper platform 52.

[0098] At the bottom of the detection box 9, that is, below the box bottom 92, a ranging radar 921 is provided. The ranging radar 921 is used to detect the distance between the bottom of the detection box 9 and underwater obstacles, and transmit the detected signal data to the main controller 31 in the form of underwater acoustic communication through the auxiliary controller 96. The main controller 31 can control the lowering speed and position of the detection box 9 according to the distance between the bottom of the detection box 9 and underwater obstacles.

[0099] It should be noted that both the main controller 31 and the auxiliary controller 96 are configured with wireless communication modules. For the wireless communication module of the main controller 31, a main antenna 2 is configured, and for the wireless communication module of the auxiliary controller 96, an auxiliary antenna 911 is configured. In this way, the main controller 31 can perform data interaction with the user terminal wirelessly, and the main controller 31 can also achieve wireless data communication with the auxiliary controller 96. When the main controller 31 receives a command issued by the user terminal wirelessly, the main controller 31 can send a control instruction to the auxiliary controller 96 wirelessly, and the auxiliary controller 96 can transmit the detection data of the water quality sensor 98 to the main controller 31 wirelessly, and the main controller 31 uploads it to the user terminal wirelessly.

[0100] The main controller 31 and the user terminal adopt a 5G communication method. When the detection box 9 is in the water body, the communication method between the main controller 31 and the auxiliary controller 96 adopts underwater acoustic communication. When the detection box 9 leaves the water body, the communication method between the main controller 31 and the auxiliary controller 96 adopts Lora communication.

[0101] The monitoring device of this embodiment operates as follows:

[0102] The monitoring device of this embodiment is used to monitor the water quality of a certain water area and obtain water quality-related data. In the target monitoring water area, the monitoring device of this embodiment is placed on the water surface. The main controller 31 controls the thruster assembly 7 to operate, so that the monitoring device moves to the target monitoring point. The main electromagnet 521 is controlled to lose power, and the detection box 9 then detaches from the detection body mooring part of the upper platform 52. The main winch 32 is controlled to lower the detection box 9. When the detection box 9 is lowered to the target depth, the main controller 31 sends a "start detection instruction" to the auxiliary controller 96 through wireless communication. The auxiliary controller 96 then starts to detect the water quality of the water body through the water quality sensor 98, and then transmits the detection data back to the main controller 31 through wireless communication. After the detection operation is completed, the main controller 31 controls the main winch 32 to start winding and recovering the detection box 9. When the detection box 9 is recovered to the detection body mooring part of the upper platform 52 and touches the limit switch 523, the main winch 32 stops the winding action. The main controller 31 controls the main electromagnet 521 to be powered on, and the mutual attraction between the main electromagnet 521 and the passive magnet 95 causes the detection box 9 to be firmly adsorbed at the detection body mooring part, thus achieving stable mooring.

[0103] Embodiment 1 of the present invention also provides an intelligent water body environment monitoring method, which is implemented by using the monitoring device as described above. This monitoring method is based on the water surface image captured by the camera 4, and a water surface floating object recognition model is used to recognize the water surface floating objects.

[0104] Specifically,

[0105] 1) Floating object monitoring process

[0106] The camera 4 acquires the image information of the water surface, providing the original data for subsequent processing. The collected images are preliminarily processed, including operations such as denoising, enhancing contrast, and adjusting resolution, in order to improve the image quality and make the subsequent object detection more accurate and efficient. The object detection model is used to identify the floating objects on the water surface in the preprocessed image, and determine information such as their positions, sizes, and shapes. The results of the object detection are deeply analyzed, and the quantity, types, distribution, etc. of the floating objects are counted. The data obtained from the analysis is uploaded to the user-side platform so that the monitoring results can be obtained and viewed in a timely manner.

[0107] 2) Model improvement for floating objects on the water surface

[0108] The floating object recognition model includes three modules, namely the feature extraction network Backbone module, the feature fusion network Neck module, and the feature recognition network Head module.

[0109] Specifically, referring to Figures 18 to 20 , in this embodiment, in the core architecture of the floating object recognition model, two-dimensional wavelet convolution Wavelet Convolution (WTConv2d) is creatively introduced, aiming to enhance the model's ability to capture global information of the image and reduce the model's parameter quantity and computational complexity.

[0110] Two-dimensional wavelet convolution can be regarded as integrating the idea of wavelet transform into the convolution operation. Let the input feature Figure X ∈R H×W×Cin , where H and W are the height and width of the feature map respectively, and C in is the number of input channels. The wavelet convolution kernel K ∈ R Kh ×Kw×Cin×Cout , where K h and K w are the height and width of the convolution kernel, and C out is the number of output channels. The output feature map Y ∈ R H1×W1×Cout of the two-dimensional wavelet convolution can be calculated by the following formula:

[0111]

[0112] Where:

[0113] Y i,j,Cout represents the value of the output feature map at the position (i, j) and channel C out .

[0114] represents the input feature map at the position (i + m, j + n) and channel Cin value

[0115] represents the weight of the wavelet convolution kernel at position (m,n) from input channel C in to output channel C out weight

[0116] The C3K2_WTConv module of the feature extraction network Backbone module described above innovatively integrates two-dimensional wavelet convolution (WTConv2d) on the basis of the C3K2 module of YOLOv11, and replaces the original Bottleneck module with a Bottleneck_WTConv module integrated with two-dimensional wavelet convolution. Determine the state of C3k according to the number of channels, and dynamically select Bottleneck_WTConv or ordinary convolution Conv. This design can flexibly adjust the feature extraction method according to the specific network structure and task requirements. If the number of channels meets specific conditions, use a convolutional layer combined with wavelet transform, otherwise use an ordinary convolutional layer.

[0117] For the Bottleneck_WTConv module described above, without using a residual connection network, the number of input channels c1 is reduced to obtain the number of channels c_ of the intermediate layer. If the number of channels c_ of the intermediate layer is equal to the number of output channels c2, WTConv2d will be selected to further process the features; if not, the conv convolution method will be used to adjust the number of channels and features. By setting the number of channels c_ of the intermediate layer, the Bottleneck_WTConv module can flexibly adjust the dimension of the intermediate layer according to different input and output channel number requirements. The Bottleneck_WTConv module helps to improve the performance and efficiency of the entire network model by combining the characteristics of wavelet convolution.

[0118] The WTConv2d is a convolution combined with wavelet transform: First, the input data first passes through the CWFilter (CreateWavelet Filter) to create a wavelet filter module to generate a filter, and then enters the WTProc (Wavelet TransformProcess) wavelet transform processing module (with a stride of 2) for wavelet transform. Subsequently, it enters the MLWT (Multi-LevelWavelet Transform) multi-level wavelet transform module to perform multi-level wavelet transform operations on the input data (the number of loops is determined by wt_levels). If the size of the input data is not even, it is first padded, and then the WTProc (Wavelet TransformProcess) wavelet transform is used to obtain the wavelet coefficients of four sub-bands. The low-frequency part is extracted and reshaped, convolved, and scaled, and then the low-frequency x_ll and high-frequency x_h parts are stored. Conv2d two-dimensional convolution and Scale scaling operations are performed on the stored data. After coming out of the MLWT (Multi-Level Wavelet Transform) module, a part of the data is scaled by the Base_scale basic scaling module after passing through the Base_conv basic convolution layer, and another part enters the InvWT (Inverse Wavelet Transform) inverse wavelet transform module. Finally, the results of the basic convolution scaling and the inverse wavelet transform are added together. If the stride is greater than 1, it also needs to go through the do_stride process to obtain the output. This process combines wavelet transform and convolution, leveraging the advantages of wavelet transform to capture different frequency features and traditional convolution to extract spatial features to obtain a better feature representation.

[0119] The Neck module of the feature fusion network adds a triple attention mechanism module Triplet Attention on the basis of the Neck module of YOLOv11. It is located in the middle and rear part of the network structure, right before the C3k2_WTConv(False) module after upsampling (Upsample) and feature concatenation (Concat). It helps to improve the detection accuracy of small targets on the water surface such as plastic debris, leaves, and water bottles.

[0120] 3) Detection effect of the floating object recognition model on the water surface

[0121] The model evaluation parameters include Precision, Recall, Average Precision (AP), mean Average Precision (mAP), and the number of model parameters (parameters), etc.

[0122] Table 1 shows the experimental effects of using different networks to train the floating object detection model on the water surface

[0123] Table 1 Experimental Results of Different Networks

[0124] Network Precision(%) Recall(%) mAP(50 - 90%) parameters GFLOPs Our_model 0.963 0.944 0.767 2471979 6.2 YOLOv11 0.951 0.939 0.765 2582347 6.3 YOLOv10 0.886 0.855 0.672 2694806 8.2 YOLOv8 0.946 0.903 0.703 2684563 6.8

[0125] Figure 21 The P-R curves of the proposed network and comparison networks (YOLOv11, YOLOv10, and YOLOv8) are shown. The test data presented in Table 1 was collected from a self-collected dataset of floating objects on the water surface and was tested by extracting data at a ratio of 20% to form a test set.

[0126] Specifically, the average precision (AP) of the four network models is 97.7% for our_model, 97.3% for YOLOv11, 91.9% for YOLOv10, and 94.9% for YOLOv8. The AP of the our_model network model is 0.4%, 5.8%, and 2.8% higher than that of YOLOv11, YOLOv10, and YOLOv8, respectively.

[0127] The precision of the four network models is 96.3% for our_model, 95.1% for YOLOv11, 88.6% for YOLOv10, and 94.6% for YOLOv8. The precision of the our_model network model is 1.2%, 7.7%, and 1.7% higher than that of YOLOv11, YOLOv10, and YOLOv8, respectively.

[0128] The recall of the four network models is 94.4% for our_model, 93.9% for YOLOv11, 85.5% for YOLOv10, and 90.3% for YOLOv8. The recall of the our_model network model is 0.5%, 8.9%, and 4.1% higher than that of YOLOv11, YOLOv10, and YOLOv8, respectively.

[0129] The mean average precision (mAP50-90) of the four network models is 76.7% for our_model, 76.5% for YOLOv11, 67.2% for YOLOv10, and 70.3% for YOLOv8. The mAP50-90 of the our_model network model is 0.2%, 9.5%, and 6.4% higher than that of YOLOv11, YOLOv10, and YOLOv8, respectively.

[0130] The parameter counts of the 4 network models are our_model (2,471,979), YOLOv11 (2,582,347), YOLOv10 (2,694,806), and YOLOv8 (2,684,563) respectively. The parameters of the our_model network model are 4.27%, 8.26%, and 7.91% lower than those of YOLOv11, YOLOv10, and YOLOv8 respectively.

[0131] The monitoring device and method of Embodiment 1 have the following advantages:

[0132] 1) Mobility and flexibility are improved. The unique underwater thruster layout enables omnidirectional flexible movement and stable operation, expanding the monitoring range.

[0133] 2) Cost is reduced and maintenance is convenient. The detection box 9 is separated from the platform base 5 and transmits data wirelessly, reducing cable damage and maintenance costs.

[0134] 3) Efficient wireless charging is achieved. The structural settings of the main electromagnet 521 and the passive magnet 95 solve the problem of aligning the detection box 9 with the coil of the upper platform 52, improving charging efficiency and safety, and ensuring stable energy supply for the detection box.

[0135] 4) The recovery positioning is accurate. The setting of the limit switch 523 helps to accurately judge whether the detection box 9 reaches the berthing part of the detection body during recovery and controls the main winch 32 to stop winding, improving the recovery accuracy and reliability, and reducing equipment damage and the uncertainty of manual intervention.

[0136] 5) Automatic precise control is realized. The ranging radar 921 is set, so that the detection box 9 can automatically and accurately sense the distance from obstacles when lowering and regulate the servo motor, improving safety and accuracy and reducing the collision risk.

[0137] 6) The reliability of unmanned operation is enhanced. The automatic design of lowering and recovery reduces the dependence on manpower, reduces the probability of human error and failure, is conducive to stable operation in complex underwater environments, expands the application range and adaptability, and provides strong support for underwater unmanned detection tasks.

[0138] 7) The contact-type water quality sensor 98 analyzes the water quality at different underwater depths layer by layer. And to ensure data stability and multi-dimensionality, the data of the non-contact water quality detector 1 is integrated to provide more comprehensive and accurate water quality information.

[0139] 8) Multiple communication methods are adopted, with strong adaptability, which can cope with complex environments; improve the stability of data transmission, complement the shortcomings; meet diverse data transmission requirements, adapt to different data; improve the reliability and flexibility of the system, with multi-link guarantee and flexible resource allocation.

[0140] 9) By means of two-dimensional wavelet convolution (WTConv2d) technology, to obtain a better feature representation, reduce the number of parameters and computational complexity of the model by 5%, and reduce the computational complexity of the water surface target detection task.

[0141] 10) Introduce the Triplet Attention mechanism in the small target detection layer so that the model can build the mutual dependence between channels and spatial positions, and improve the model's ability to understand the features of small targets.

[0142] The wireless charging receiving coil 94, passive magnet 95, auxiliary controller 96 and auxiliary lithium battery 97 are encapsulated in the box cover 93 to prevent damage caused by contact with water.

[0143] It should be noted that in this embodiment, both the main controller 31 and the auxiliary controller 96 adopt MCU controllers.

[0144] Embodiment 2:

[0145] See Figures 6 to 22 , this Embodiment 2 provides an intelligent water environment monitoring device for monitoring the water quality of a water area. This Embodiment 2 reflects the original concept of the present invention.

[0146] The monitoring device of this Embodiment 2 focuses on the fields of water quality detection and computer vision. For different water bodies such as aquaculture water, surface water, marine environment and agricultural irrigation water quality, multiple indicators such as water temperature, dissolved oxygen, pH value, COD, BOD, ammonia nitrogen, heavy metals, salinity, salt content, harmful substances, etc. are monitored respectively to achieve the purposes of water quality regulation, ecological protection, ensuring water use safety and healthy growth of crops.

[0147] Regarding the monitoring of water surface floating objects, a new water surface target detection method is proposed. Significantly improve the detection performance of specific targets such as water surface floating objects, and achieve accurate identification and efficient positioning of floating objects in complex water surface environments.

[0148] The monitoring device of this Embodiment 2 monitors multiple indicators of different water bodies, can accurately grasp the water quality status, provide real-time data for water quality regulation and intelligent operation and maintenance, and the water surface target detection model adopted can reduce the computational complexity of the target detection task and improve the accuracy of target detection.

[0149] The monitoring device of this Embodiment 2 mainly includes the following system components: water surface motion control system, detection box position control system, multi-dimensional water quality monitoring system, water surface floating object monitoring system.

[0150] I. Water Surface Motion Control System

[0151] 1. Motion Structure Design

[0152] As Figure 8 , 9 shown, eight underwater thrusters (8, 10, 11, 12, 21, 48, 49, 50) are installed below the device. Thrusters (11, 42) are installed at both ends of the edge of the device platform, thrusters (10, 41) are installed on both sides of the floating body of the device, and thrusters (8, 12, 21, 43) are arranged in a square layout, with their propulsion directions perpendicular to the downward direction and installed at the bottom of the device. Compared with the existing water quality monitoring devices that mostly adopt fixed buoy designs, this design can achieve the all-round flexible movement of the device and effectively reduce the sway amplitude of the device.

[0153] 2. Attitude and Position Control

[0154] The device is equipped with a gyroscope and a GPS satellite navigation system (23) as Figure 10 shown. It can reduce the sway amplitude of the platform during the operation of the device and ensure accurate reading of the position and attitude data of the device.

[0155] The control process is as Figure 14 shown: During the operation, the MCU controller reads the attitude angle of the device through the gyroscope, reads the position of the device through the GPS satellite navigation system, combines with the expected attitude value of the operating state of the device, and after performing PID operations, obtains the parameters that need to be adjusted by the electric drive to control the output of the thrusters.

[0156] 3. Fixing the Water Surface Position

[0157] Structurally: As Figure 6 shown, the device platform is equipped with a hinge motor (4), an anchor (7), and a pulley (6). The chain of the hinge motor is connected to the anchor through the pulley.

[0158] The control process is as follows:

[0159] 1) When the position of the device needs to be fixed, the electric drive controller controls the hinge motor to lower the anchor. When the anchor touches the bottom of the water, the load of the hinge motor is reduced, and the motor current value decreases significantly. At this time, the anchor is in the deployed state, and the hinge motor stops lowering the anchor.

[0160] 2) When the position of the device needs to be moved, the electric drive controller controls the hinge motor to retract the anchor upward. When the anchor is retracted to the platform, the hinge motor is blocked, and the motor current value increases. At this time, the anchor is in the retracted state, and the hinge motor stops retracting the anchor upward.

[0161] II. Detection Box Position Control System

[0162] 1. Detection Box Retraction Design

[0163] Structurally: As Figure 8The figure shows a schematic diagram after the device is fixed for recovery. The detection box is stored between the two platforms of the device, and there is an unobstructed design around it. And as Figure 12 shown, an electromagnet (28), a wireless charging transmitting coil (29) and a limit switch (27) are installed under the device platform, and a magnet (34) and a wireless charging receiving coil (36) are installed on the top of the detection box. This structure enables the detection box to be accurately stored between the two platforms. While wirelessly charging the lithium battery in the box, the attachment of underwater algae, shellfish and microorganisms to the surface of the sensor is alleviated by sunlight irradiation around, reducing the maintenance cycle and cost.

[0164] The recovery process is as follows: When the hinge motor drives the detection box to recover, the top of the detection box touches the limit switch. At this time, the hinge motor stops recovering the detection box, the electromagnet above the device is energized to attract the magnet of the detection box, and the wireless charging coil charges the battery in the box.

[0165] 2. Design of the detection box dropping

[0166] Structurally: As Figure 7 、 13 shown, a servo motor (26) and a pulley (37) are installed at the center of the top of the device, which are connected to the detection box (9) through a chain (15). And an underwater radar rangefinder (31) is installed at the bottom of the detection box to ensure that the bottom of the detection box will not collide with obstacles during the falling process.

[0167] The dropping process is as follows: The MCU controller receives the dropping instruction, controls the electromagnet to cut off the power and release the detection box, controls the servo motor to drop the detection box into the water. During the falling process, the radar rangefinder measures the underwater distance in real time and sends data to the MCU controller through underwater acoustic communication to control the operation of the servo motor.

[0168] III. Multi-dimensional water quality monitoring system

[0169] 1. Water quality detection unit

[0170] The water quality detection of the device has two detection methods: contact type and non-contact type. In terms of contact type detection, as Figure 11 shown, a water quality detection box (9) that can be lifted is equipped in the center of the device. Inside the detection box, as Figure 12 、 13 shown, a variety of water quality sensors are arranged to detect water quality indicators such as water temperature, dissolved oxygen, pH value, etc. at different depths. In terms of non-contact type detection, a hyperspectral water quality detector (1) is installed at the top of the device. By analyzing the spectral characteristics of reflected light or transmitted light, parameters such as the concentration of suspended solids in water, heavy metal ions, organic matter content and nutrient salt content are detected and analyzed. Combining the two detection data, the comprehensive analysis of water quality is realized.

[0171] 2. Wireless data transmission unit

[0172] The data communication end of the detection box adopts a dual-mode data transmission method, and data is transmitted between the device platform end and the user end through 5G communication technology. The transmission process is as follows Figure 15 as shown

[0173] The user end issues a water quality monitoring command to the device platform through 5G communication.

[0174] When detecting water quality, the communication method between the detection box and the monitoring platform adopts underwater acoustic communication, which is responsible for sending the information detected by the underwater radar to the platform controller when the detection box is underwater (such as Figure 6 the state shown).

[0175] After the detection box is recovered, the communication method between it and the monitoring platform adopts Lora communication, which is responsible for transmitting the water quality parameter information detected this time to the monitoring platform when the detection box is in the recovered state (such as Figure 8 the state shown). Give full play to the advantages of the two communication methods in different environments and improve the stability of data transmission.

[0176] After the monitoring platform end receives the data sent by the detection box, it sends the water quality data collected by the device and the state data of the device to the user end through 5G communication technology.

[0177] Surface Floating Object Monitoring System

[0178] 1. Floating Object Monitoring Process

[0179] As Figure 16 shown, the image acquisition sensor obtains the image information of the water surface, providing the original data for subsequent processing. The collected images are preliminarily processed, including operations such as denoising, enhancing contrast, and adjusting resolution, so as to improve the image quality and make the subsequent object detection more accurate and efficient. The object detection model is used to identify the surface floating objects in the preprocessed images and determine their position, size, shape and other information. The results of object detection are deeply analyzed to count the number, type, distribution, etc. of the floating objects. The data obtained from the analysis is uploaded to the user end platform so that the monitoring results can be obtained and viewed in a timely manner.

[0180] 2. Model Improvement for Surface Floating Objects

[0181] As Figure 17 shown, the model consists of three modules: the feature extraction network Backbone module, the feature fusion network Neck module, and the feature recognition network Head module.

[0182] The C3K2_WTConv module of the feature extraction network Backbone module described above has a structure as Figure 18As shown in the figure, based on the C3K2 module of YOLOv11, two-dimensional wavelet convolution (WTConv2d) is innovatively integrated, and the original Bottleneck module is replaced with a Bottleneck_WTConv module that integrates two-dimensional wavelet convolution. Determine the state of C3k according to the number of channels, and dynamically select Bottleneck_WTConv or ordinary convolution Conv. This design can flexibly adjust the feature extraction method according to the specific network structure and task requirements. If the number of channels meets specific conditions, use a convolutional layer combined with wavelet transform, otherwise use an ordinary convolutional layer.

[0183] The described Bottleneck_WTConv module has a structure as Figure 19 As shown in the figure, without using a residual connection network, the number of input channels c1 is reduced to obtain the number of channels c_ of the intermediate layer. If the number of channels c_ of the intermediate layer is equal to the number of output channels c2, WTConv2d will be selected to further process the features; if not, the conv convolution method will be used to adjust the number of channels and features. By setting the number of channels c_ of the intermediate layer, the Bottleneck_WTConv module can flexibly adjust the dimension of the intermediate layer according to different input and output channel number requirements. The Bottleneck_WTConv module helps to improve the performance and efficiency of the entire network model by combining the characteristics of wavelet convolution.

[0184] The described WTConv2d is a convolution combined with wavelet transform, and its structure is as Figure 20As shown below: First, the input data first passes through the CWFilter (Create Wavelet Filter) module to create a wavelet filter, and then enters the WTProc (Wavelet Transform Process) module (with a step size of 2) for wavelet transform. Subsequently, it enters the MLWT (Multi-Level Wavelet Transform) module to perform multi-level wavelet transform operations on the input data (the number of loops is determined by wt_levels). If the size of the input data is not even, it is first padded, and then the WTProc (Wavelet Transform Process) module is used to obtain the wavelet coefficients of four subbands through wavelet transform. The low-frequency part is extracted and reshaped, convolved, and scaled, and then the low-frequency x_ll and high-frequency x_h parts are stored. Conv2d two-dimensional convolution and Scale scaling operations are performed on the stored data. After coming out of the MLWT (Multi-Level Wavelet Transform) module, a part of the data is scaled by the Base_scale module after passing through the Base_conv basic convolution layer, and another part enters the InvWT (Inverse Wavelet Transform) module. Finally, the results of the basic convolution scaling and the inverse wavelet transform are added together. If the step size is greater than 1, it is also necessary to go through the do_stride process to obtain the output. This process combines wavelet transform and convolution, taking advantage of the ability of wavelet transform to capture different frequency features and the ability of traditional convolution to extract spatial features to obtain a better feature representation.

[0185] The described feature fusion network Neck module is as Figure 17 shown. Based on the Neck module of YOLOv11, a triple attention mechanism module Triplet Attention is added, which is located in the middle and rear part of the network structure, right before the C3k2_WTConv (False) module after upsampling (Upsample) and feature concatenation (Concat). This helps to improve the detection accuracy of small targets on the water surface such as plastic debris, leaves, and water bottles.

[0186] 3. Model Parameter Evaluation

[0187] The model evaluation parameters of the monitoring device in Embodiment 2 include precision, recall, average precision (AP), mean average precision (mAP), and the number of model parameters (parameters), etc.

[0188] 1) Precision is the proportion of true positives (TP) among all samples predicted as positive. The formula is:

[0189]

[0190] where TP is the number of true positives, that is, the number of samples correctly predicted as positive by the model; FP is the number of false positives, that is, the number of samples that the model wrongly predicts the negative class as positive.

[0191] 2) Recall is the proportion of samples correctly predicted as positive by the model among all true positive samples. The formula is:

[0192]

[0193] where FN is the number of false negatives, that is, the number of samples that the model wrongly predicts the positive class as negative.

[0194] 3) Average Precision (AP) is the average of precisions at different recall levels, which is used to measure the performance of the model on a certain class. For the P-R (Precision-Recall) curve, AP is the area under the curve.

[0195] 4) Mean Average Precision (mAP) When the model is used for multi-class object detection, mAP is the average of APs for all classes, which comprehensively evaluates the detection performance of the model on multiple classes. The formula is:

[0196]

[0197] 5) The number of model parameters is the number of learnable parameters in the model, and it is an index to measure the computational complexity of the model, the same as GFLOPs. The number of model parameters directly affects the computational amount and storage requirements of the model. Too many parameters may lead to a slow inference speed of the model and cannot meet the requirements of real-time detection.

[0198] Table 1 shows the experimental results of training the water surface floating object detection model using different networks. Figure 21 The P-R curves of the proposed network and comparison networks (YOLOv11, YOLOv10, and YOLOv8) are shown. The test data presented is from the self-collected water surface floating object dataset, and a test set is formed by extracting 20% of the data for testing.

[0199] Specifically, the average precision (AP) of the four network models is 97.7% for our_model, 97.3% for YOLOv11, 91.9% for YOLOv10, and 94.9% for YOLOv8. The AP of our_model is 0.4%, 5.8%, and 2.8% higher than that of YOLOv11, YOLOv10, and YOLOv8 respectively.

[0200] The precision of the four network models is 96.3% for our_model, 95.1% for YOLOv11, 88.6% for YOLOv10, and 94.6% for YOLOv8. The precision of our_model is 1.2%, 7.7%, and 1.7% higher than that of YOLOv11, YOLOv10, and YOLOv8 respectively.

[0201] The recall of the four network models is 94.4% for our_model, 93.9% for YOLOv11, 85.5% for YOLOv10, and 90.3% for YOLOv8. The recall of our_model is 0.5%, 8.9%, and 4.1% higher than that of YOLOv11, YOLOv10, and YOLOv8 respectively.

[0202] The mean average precision (mAP50-90) of the four network models is 76.7% for our_model, 76.5% for YOLOv11, 67.2% for YOLOv10, and 70.3% for YOLOv8. The mAP50-90 of our_model is 0.2%, 9.5%, and 6.4% higher than that of YOLOv11, YOLOv10, and YOLOv8 respectively.

[0203] The number of parameters of the four network models is 2,471,979 for our_model, 2,582,347 for YOLOv11, 2,694,806 for YOLOv10, and 2,684,563 for YOLOv8. The parameters of our_model are 4.27%, 8.26%, and 7.91% lower than those of YOLOv11, YOLOv10, and YOLOv8 respectively.

[0204] The GFLOPs of the four network models are 6.2 for our_model, 6.3 for YOLOv11, 8.2 for YOLOv10, and 6.8 for YOLOv8. The GFLOPs of our_model are 0.1, 2, and 0.6 lower than those of YOLOv11, YOLOv10, and YOLOv8 respectively.

[0205] Based on the values of average precision (AP), precision, recall, the number of model parameters (parameters), and GFLOPs, it is proven that the proposed model is superior to the compared networks. It can be concluded that the proposed model may have more advantages than YOLOv11, YOLOv10, and YOLOv8.

[0206] Figure 20 For the actual detection effect diagram of the model, as Figure 20 shown, it contains multiple detection results of floating objects on the water surface. The monitoring device of Embodiment 2 can effectively detect the floating object targets on the water surface.

[0207] The following provides in Table 2 Figures 6 to 13 the interpretations of the reference numerals in

[0208] Table 2, Annotations of the Water Environment Monitoring Device

[0209]

[0210]

[0211] Embodiment 2 proposes a mobile multi-dimensional water environment intelligent monitoring device, which solves the water quality monitoring requirements for different water depths in a large range by arranging the depth of controllable sensors, enhancing the stability of fixed-point sampling, improving mobility and flexibility, and enhancing data transmission stability. In addition, the sensor can be lifted above the water surface, and the attachment of underwater algae, shellfish, and microorganisms to the sensor surface can be alleviated by sunlight irradiation, improving the accuracy of data monitoring and reducing the maintenance cycle and cost. For the visual monitoring of floating objects on the water surface, a water surface target detection method based on the improved YOLOv11 is proposed. The object detection model adopted by this method is constructed based on the improved YOLOv11 model. The C3K2 module in its Backbone module innovatively integrates two-dimensional wavelet convolution (WTConv2d), and the Triplet Attention attention mechanism is embedded in the Neck module. With the two-dimensional wavelet convolution (WTConv2d), a more efficient, robust, and easily integrated convolutional layer solution is provided to obtain better feature representations, and the model can reduce the number of parameters and computational complexity by 10%, which can reduce the computational complexity of the water surface target detection task. Introducing the Triplet Attention attention mechanism in the small target detection layer enables the model to construct the mutual dependence between channels and spatial positions, improving the model's understanding ability of small target features. It improves the accuracy of water surface target detection and better meets the requirements of floating object recognition on the water surface.

[0212] The main innovation of the monitoring device of Embodiment 2 lies in:

[0213] 1) The 8 underwater thrusters placed at the bottom of the device adopt a unique structural layout, combined with a gyroscope and GPS navigation system, to reduce its shaking during movement and expand the coverage of water quality detection.

[0214] 2) A servo motor is used in conjunction with an underwater radar rangefinder to control the placement of the detection box, so that the detection box can stably analyze the water quality at different depths.

[0215] 3) The detection box is hinged to the device, and the detection box can be lifted up to the inside of the device for storage. The surroundings are unobstructed. The sunlight is fully utilized to solve the problem of underwater biological attachment that has long plagued water quality detection devices, effectively reducing the complexity and cost of maintenance.

[0216] 4) Combining the two water quality detection methods of contact water quality sensor and non-contact hyperspectral water quality detector, the types of detection parameters are increased, and comprehensive analysis of multi-dimensional parameters of water bodies is achieved.

[0217] 5) For visual detection of floating objects on the water surface, a network model based on improved YOLOv11 is proposed, replacing the original C3K2 module of the Backbone module with the C3K2_WTConv module that integrates two-dimensional wavelet convolution (WTConv2d). The Triplet Attention mechanism is introduced in the small target detection layer of the Neck module.

[0218] The monitoring device of this embodiment 2 can achieve the following innovative effects:

[0219] 1) Improved maneuverability and flexibility. The unique layout of 8 underwater thrusters enables all-round flexible movement and stable operation, expanding the monitoring range.

[0220] 2) Reduce costs and facilitate maintenance. The detection box is separated from the platform and transmits data wirelessly, reducing cable damage and maintenance costs.

[0221] 3) Achieve efficient wireless charging. The unique structure solves the problem of misalignment between the detection box and the platform coil, improves charging efficiency and safety, and ensures stable energy supply to the detection box.

[0222] 4) Accurate recycling positioning. The limit switch helps to accurately determine whether the recycling has reached the designated area and control the motor to stop, thereby improving the recycling accuracy and reliability and reducing equipment damage and uncertainty of manual intervention.

[0223] 5) Realize automated and precise control. The underwater radar distance detector enables the detection box to automatically and accurately sense the distance to obstacles and adjust the servo motor when it is lowered, improving safety and accuracy and reducing the risk of collision.

[0224] 6) Enhance the reliability of unmanned operation. The automated design of lowering and recovery reduces the dependence on human labor, decreases the probability of human errors and failures, facilitates stable operation in complex underwater environments, expands the application scope and adaptability, and provides strong support for underwater unmanned detection tasks.

[0225] 7) The contact sensors analyze the water quality at different depths underwater layer by layer. And to ensure the data stability and multi-dimensionality, the data of the non-contact hyperspectral imager is integrated to provide more comprehensive and accurate water quality information.

[0226] 8) The device adopts multiple communication methods, has strong adaptability, and can cope with complex environments; it improves the data transmission stability, complements the short board; meets the diverse data transmission requirements, adapts to different data; enhances the system reliability and flexibility, and ensures multiple links and can flexibly allocate resources.

[0227] 9) By means of the two-dimensional wavelet convolution (WTConv2d) technology, better feature representations can be obtained, and the model can reduce the number of parameters and the amount of computation by 5%, reducing the computational complexity of the water surface target detection task.

[0228] 10) The Triplet Attention mechanism is introduced into the small target detection layer, enabling the model to build the mutual dependence between channels and spatial positions, and improving the model's ability to understand the features of small targets.

[0229] The above are only the preferred embodiments of the present invention, and are not used to limit the protection scope of the present invention. Therefore, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent monitoring device for water environment, characterized in that: The monitoring device comprises a platform base (5), a thruster assembly (7), a water quality detector (1), a camera (4), a detection box (9), a main winch (32) and a main controller (31); The propeller assembly (7), the water quality detector (1), the camera (4), the main winch (32) and the main controller (31) are all installed and arranged based on the platform base (5); The detection box (9) is assembled and connected to the hoisting cable (321) of the main hoist (32), and the main hoist (32) is capable of lowering and hoisting the detection box (9); The propeller assembly (7) is used to drive the monitoring device to move autonomously on the water surface.

2. The water environment intelligent monitoring device according to claim 1 is characterized in that: The detection box (9) is provided with a water quality sensor (98), an auxiliary controller (96) and an auxiliary lithium battery (97).

3. The water environment intelligent monitoring device according to claim 2 is characterized in that: The main controller (31) and the auxiliary controller (96) are both equipped with a wireless communication module, and wireless data communication can be performed between the main controller (31) and the auxiliary controller (96). The main controller (31) can send control instructions to the auxiliary controller (96) by wireless means, and the auxiliary controller (96) can transmit detection data of the water quality sensor (98) to the main controller (31) by wireless means.

4. The intelligent water environment monitoring device according to claim 3 is characterized in that: When the detection box (9) is in a body of water, the communication method between the main controller (31) and the auxiliary controller (96) is hydroacoustic communication; When the detection box (9) leaves the water body, the communication method between the main controller (31) and the auxiliary controller (96) adopts Lora communication.

5. The water environment intelligent monitoring device according to claim 1 is characterized in that: The platform base (5) is provided with a detection body docking portion, and the detection body docking portion is used for docking the detection box (9).

6. The water environment intelligent monitoring device according to claim 5 is characterized in that: A main electromagnet (521), a wireless charging transmitting coil (522) and a limit switch (523) are arranged at the docking portion of the detection body, and the detection box (9) is provided with a passive magnet (95) and a wireless charging receiving coil (94); When the detection box (9) is docked at the docking portion of the detection body, the limit switch (523) is triggered, the main electromagnet (521) is energized to generate magnetic force, the main electromagnet (521) and the passive magnet (95) attract each other, and the wireless charging transmitting coil (522) and the wireless charging receiving coil (94) are correspondingly close together, so that the detection box (9) can be positioned and docked at the docking portion of the detection body and realize wireless charging.

7. The water environment intelligent monitoring device according to claim 1 is characterized in that: A distance measuring radar (921) is arranged at the bottom of the detection box (9); The monitoring device is equipped with a gyroscope and a GPS positioning component.

8. A method for intelligent monitoring of water environment, the monitoring method using the monitoring device as claimed in claim 1; Features: The monitoring method adopts a water surface floating object recognition model to identify the water surface floating objects; The water surface floating object recognition model is constructed based on the improved YOLOv11 model; The water surface floating object recognition model includes a feature extraction network Backbone module, a feature fusion network Neck module and a feature recognition network Head module; The C3K2 module in the feature extraction network Backbone module is replaced with the C3K2_WTConv module that incorporates a two-dimensional wavelet convolution algorithm; In the C3K2_WTConv module, the original Bottleneck module is replaced with the Bottleneck_WTConv module; A two-dimensional wavelet convolution feature processing algorithm option is added to the Bottleneck_WTConv module, and whether to use the two-dimensional wavelet convolution algorithm is selected according to the number of intermediate channels after the number of input channels is reduced in dimension; The feature fusion network Neck module embeds the triple attention mechanism module TripletAttention in the small target detection layer.

9. The method for intelligent monitoring of water environment according to claim 8, characterized in that: The training process of the water surface floating object recognition model includes: Constructing a water surface floating object image dataset, wherein the water surface floating object image dataset includes a training set, a validation set, and a test set; All target objects in the water surface floating object image dataset are marked with corresponding borders, and corresponding labels for indicating the category of the water surface floating object image are set; The images of the water surface floating object training set are respectively input into the network model improved based on YOLOv11 for training, so as to obtain a trained water surface floating object recognition model; Using the water surface floating object verification set images, verifying the trained water surface floating object recognition model; The performance of the trained floating object recognition model is evaluated using the floating object test set, and the performance evaluation includes calculating accuracy, recall rate and F1 score indicators.

10. The method for intelligent monitoring of water environment according to claim 8, characterized in that: In the two-dimensional wavelet convolution feature processing algorithm, wavelet transform is combined with convolution, which combines the advantages of wavelet transform in capturing different frequency features and traditional convolution in extracting spatial features.

Images