Accumulated water depth monitoring and identifying device based on artificial intelligence

Abstract

The invention relates to the field of computer vision, and particularly discloses an artificial intelligence-based accumulated water depth monitoring and recognition device, which comprises a high-definition camera for continuously acquiring road surface image data of a monitoring area; the image processing module is used for receiving and processing image data of the high-definition camera; the image processing module comprises an image preprocessing unit and a ponding area identification unit, the image preprocessing unit is used for graying and denoising the image, and the ponding area identification unit is used for identifying the ponding range. According to the ponding depth monitoring and identifying device based on artificial intelligence, the Mask R-CNN model is adopted to accurately identify and process the floaters, an improved U-Net architecture attention mechanism is combined, the floaters are effectively prevented from interfering ponding area identification and depth calculation, the problem that the floaters are misjudged to be ponding areas in the prior art is solved, and the accuracy of ponding area identification and depth calculation is improved. And errors caused by floating objects can be accurately compensated during depth calculation.

Description

Technical Field

[0001] The present invention relates to the field of computer vision, and in particular to an artificial intelligence-based water depth monitoring and identification device. Background Art

[0002] Waterlogging depth monitoring and identification systems play a vital role in today's urban development and management. With the frequent occurrence of extreme weather, urban flooding is becoming increasingly serious. Accurate, real-time information on waterlogging depth has become a key requirement for ensuring smooth urban traffic, the safety of residents and property, and the stable operation of infrastructure. Traditional waterlogging depth monitoring methods rely primarily on simple water level sensors. These sensors typically require direct contact with the water, limiting their installation locations and only capturing single-point water depth data, making it difficult to fully capture the full picture of waterlogging across large areas. This is why image recognition-based waterlogging monitoring technology has emerged. It uses cameras to capture road surface images and employs algorithms to analyze waterlogged areas.

[0003] However, existing image recognition technology has certain defects when facing actual complex scenes, among which the impact of floating objects on the water surface becomes a major problem. In real waterlogging scenarios, there are many types of floating objects, including leaves, branches, garbage, and plastic foam. They come in various shapes and sizes and are highly random. These floating objects bring multi-faceted interference to water depth monitoring: On the one hand, during the waterlogging area identification stage, the color and texture of floating objects may be similar to those of the water, making them easily misidentified as part of the waterlogging area, or they may obscure the water boundary, resulting in inaccurate waterlogging area segmentation. For example, large pieces of floating leaves piled on the surface of the waterlogging may cause the system based on traditional semantic segmentation algorithms to mistakenly identify the leaf area as a continuous area of ​​waterlogging, when in fact the waterlogging range is not the case. This makes the subsequent depth calculation lose the accurate regional basis and has a huge error. On the other hand, the presence of floating objects also creates many obstacles for the depth calculation process. When there are floating objects on the water surface, it will change the reflective properties of the water surface. Traditional dynamic water surface processing algorithms such as optical flow cannot distinguish whether the light changes are due to water fluctuations or floating objects blocking the reflection, resulting in misjudgment of water flow velocity, and then deviations when restoring the dynamic water surface to static state for depth calculation.

[0004] Therefore, it is necessary to propose an artificial intelligence-based water depth monitoring and identification device to solve the above problems. Summary of the Invention

[0005] The main purpose of the present invention is to provide an artificial intelligence-based water depth monitoring and identification device that can effectively solve the problems in the background technology.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] An artificial intelligence-based water depth monitoring and identification device, comprising:

[0008] High-definition camera, used to continuously collect road surface image data in the monitoring area;

[0009] Image processing module, used to receive and process image data from high-definition cameras;

[0010] The image processing module includes an image preprocessing unit and a waterlogging area recognition unit. The image preprocessing unit is used to perform grayscale and denoising processing on the image. The waterlogging area recognition unit is used to identify the waterlogging range. The steps of identifying the waterlogging range include:

[0011] S1.1: Using the deep learning-based instance segmentation model Mask R-CNN, the preprocessed image is input. The model first uses a convolutional neural network to extract multi-scale feature maps of floating objects.

[0012] S1.2: A region proposal network is used to slide a window on the feature map to generate anchor boxes of different scales and aspect ratios as candidate floating object regions. A small fully connected network is used to calculate the probability score of the anchor box containing the floating object and the offset relative to the actual floating object position.

[0013] S1.3: For the candidate area after screening, determine whether it contains floating objects through the classification branch binary classification and output the classification probability P cls , the mask branch generates a pixel-level mask to outline the foreground candidate area, denoted as M;

[0014] S1.4: According to the classification probability P cls Set a threshold, determine the floating object instances, and process the original image with the mask M to remove the floating objects or mark them as special areas, and output the purified image to the next step;

[0015] S1.5: Receive the cleaned image and downsample it using a U-Net model encoder with an attention mechanism. After a series of convolution and pooling layers, the image resolution is reduced, the number of feature map channels is increased, and multi-scale water accumulation features are extracted.

[0016] S1.6: When the decoder is upsampling, the attention mechanism intervenes to calculate the weights associated with pixels and global features, highlighting water-logged pixels and suppressing the influence of floating objects and irrelevant pixels.

[0017] S1.7: The decoder fuses feature maps of different scales, restores the resolution through convolution and upsampling, and finally outputs a binary image of the waterlogged area. The cross-entropy loss function is used to optimize the model and accurately identify the waterlogged area. The cross-entropy loss function formula is:

[0018] Where N is the number of samples, y i is the true label, is the model prediction value;

[0019] It also includes a water depth calculation unit, which is used to calculate the water depth.

[0020] Preferably, the attention mechanism introduced into the U-Net architecture in step S1.5 is used to enable the model to focus on key water accumulation features when segmenting the water accumulation area and ignore the interference of floating objects.

[0021] Preferably, the water depth calculation unit includes a dynamic water surface processing unit, and the dynamic water surface processing unit is used to calculate the water depth of the water accumulation area when there are floating objects on the dynamic water surface;

[0022] The steps to calculate the water depth when there are floating objects on the dynamic water surface are:

[0023] S2.1: Use the SIFT algorithm to extract stable feature points from adjacent frames, avoiding interference from floating objects and locking in relatively stable areas on the surface of the accumulated water;

[0024] S2.2: Combined with the optical flow method, assume that the pixel values ​​of the feature points of adjacent frame images are I(x, y, t) and I(x+Δx, y+Δy, t+Δt), solve the optical flow field according to the optical flow constraint equation, compensate for the dynamic water surface and calculate the depth after approximating it to static, where the optical flow constraint equation is I x u+I y v+I t =0, where I x , I y is the image partial derivative, I t is the time partial derivative, u and v are the pixel velocity components.

[0025] Preferably, when calculating the water depth when there are floating objects on the dynamic water surface, a motion model is built based on the motion trajectory and speed of the floating objects in the previous frames, and the occlusion effect of the current frame on the calculation of the water depth is predicted. The occluded area is dynamically adjusted or the suspicious area is marked for verification by multiple frames of data.

[0026] Preferably, the water depth calculation unit further includes a slope correction unit, and the slope correction unit is used to calculate the water depth when the water accumulation area is a slope and has floating objects;

[0027] The method for calculating the water depth when the waterlogged area is a slope and there are floating objects is as follows: using the known reference object in the image, set the camera optical center O, the reference object projection point A, the waterlogged point P, the known actual height H of the reference object, and the pixel height h A , water point pixel height h P , according to the principle of similar triangles, the depth of water is

[0028] Preferably, when calculating the water depth when the waterlogged area is a slope and there are floating objects, the water depth data around the reference object when there are no floating objects is compared with the current frame data based on the influence of reflection and occlusion of the floating objects, and the water depth of the affected area is corrected using statistical methods.

[0029] Preferably, it also includes a data transmission module, which is used to upload the water depth data and image information calculated by the image processing module to the cloud server or local monitoring center in real time.

[0030] Preferably, in step S1.1, the multi-scale features include but are not limited to the shape, texture, and color information of the floating object.

[0031] Preferably, it also includes a self-learning optimization module, which is used to regularly compare the collected images, the calculated water depth results and the actual manual measurement data. If there is a large deviation, the deep learning model parameter weights are automatically adjusted.

[0032] Preferably, an auxiliary lighting system is also included, which includes an intelligent light sensor for detecting light intensity and an LED lamp for adding a light source to the high-definition camera. A threshold is preset based on the ambient light intensity. When the light is dark below the threshold, the LED lamp is automatically triggered for lighting.

[0033] Compared with the existing technology, the present invention provides an artificial intelligence-based water depth monitoring and identification device, which has the following beneficial effects:

[0034] 1. This AI-based water depth monitoring and identification device uses the Mask R-CNN model to accurately identify and process floating objects. Combined with the improved U-Net architecture attention mechanism, it effectively prevents floating objects from interfering with water area identification and depth calculation. This not only overcomes the problem of traditional technologies misjudging floating objects as water areas, but also accurately compensates for the errors caused by floating objects during depth calculation, ensuring that the accuracy of water depth monitoring is greatly improved.

[0035] 2. This artificial intelligence-based water depth monitoring and identification device uses the SIFT algorithm combined with the optical flow method to process dynamic water surfaces, extracts stable feature points to avoid interference from floating objects, and accurately solves the optical flow field to restore an approximately static water surface. It overcomes the shortcomings of traditional optical flow methods that are confused by reflections from floating objects and water flow fluctuations, provides a reliable basis for calculating water depth, and adapts to complex dynamic water environments such as turbulent water flows and surface ripples.

[0036] 3. This artificial intelligence-based water depth monitoring and identification device uses known reference objects in images and corrects the impact of floating objects based on the principle of similar triangles and statistical methods. It fully considers the reflection of floating objects on the surface of the water and visual occlusion factors, accurately compensates for the depth error caused by the slope surface, and solves the problem of large deviation in traditional technology in measuring water depth on slope terrain.

[0037] 4. This AI-based water depth monitoring and identification device, in terms of hardware, uses high-definition cameras, intelligent auxiliary lighting, and solar power supply to ensure stable image acquisition around the clock; in terms of software, the self-learning optimization module continuously improves the accuracy and adaptability of the system, and all links work closely together. Compared with the existing reliance on single-point water level sensors or simple image recognition technology, it can accurately monitor the depth of water accumulation in all directions and in real time, providing strong support for urban management and emergency decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a structural schematic diagram of the present invention;

[0039] Figure 2 It is a structural diagram of the image processing module of the present invention;

[0040] Figure 3 It is a structural schematic diagram of a water depth calculation unit of the present invention;

[0041] Figure 4 It is a structural schematic diagram of the auxiliary lighting system of the present invention. DETAILED DESCRIPTION

[0042] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0043] like Figures 1-4 As shown, an artificial intelligence-based water depth monitoring and identification device includes a high-definition camera. A suitable location is selected in the target monitoring area, and the high-definition camera is installed. According to the monitoring range and scene characteristics, the camera is adjusted to an appropriate height and angle to ensure that it can clearly capture road surface images under different lighting conditions and continuously collect road surface image data in the monitoring area.

[0044] Furthermore, the system also includes an image processing module. The image processing module can be an embedded device equipped with a high-performance GPU and has strong real-time computing capabilities. Its function is to receive and process camera image data, run a series of deep learning models, and provide computing power support for the entire system. The image processing module includes an image preprocessing unit and a water accumulation area recognition unit. The image preprocessing unit is used to grayscale and denoise the image.

[0045] When graying, the formula Gray = 0.299*R + 0.587*G + 0.114*B is used to convert the color image into a grayscale image, reduce the data dimension, retain the basic texture features, and reduce the subsequent calculation amount. R, G, and B are the red, green, and blue channel values ​​of the color image pixel respectively, and Gray is the grayscale value.

[0046] When denoising, Gaussian filtering is used to remove image noise. For each pixel (x, y) in the image, the formula for calculating the pixel value after filtering is:

[0047] Where I(x+i,y+i) is the original image pixel value, σ is the standard deviation of the Gaussian kernel, and k determines the filter window size, which is 1-3. This helps smooth the image and reduce the interference of noise on subsequent waterlogging identification.

[0048] The water accumulation area identification unit is used to identify the water accumulation range, and the steps of identifying the water accumulation range include:

[0049] Step 1: Use Mask R-CNN, a deep learning-based instance segmentation model, to input the preprocessed image. The model first uses a convolutional neural network to extract multi-scale feature maps, preferably a ResNet series backbone network, covering information such as the shape, texture, and color of floating objects;

[0050] Step 2: The region proposal network slides a window on the feature map to generate anchor boxes of different scales and aspect ratios as candidate floating object regions, preferably [(32, 32), (64, 64), (128, 128), (256, 256), (512, 512)] and [0.5, 1, 2] aspect ratios. A small fully connected network is used to calculate the probability score of the anchor box containing the floating object and the offset relative to the actual floating object position;

[0051] Step 3: For the candidate area after screening, the classification branch determines whether it contains floating objects and outputs the classification probability P cls ,The mask branch generates a pixel-level mask of the foreground candidate area, where the mask value 1 represents the floating object pixel and 0 represents the background pixel, accurately outlining the contour, which is recorded as M;

[0052] Step 4: According to the classification probability P cls Set a threshold, determine the floating object instances, and process the original image in combination with the mask M to remove the floating objects or mark them as special areas. When removing the floating objects, set the pixels to the background value and output the purified image to the next step;

[0053] It should be noted that the algorithm formulas designed in the above steps include anchor box and target box regression and loss function. The existing algorithm formulas are used and will not be elaborated on in detail.

[0054] Step 5: Receive the cleaned image and downsample it using a U-Net model encoder with an attention mechanism. The attention mechanism enables the model to focus on key water features when segmenting waterlogged areas, ignoring interference from floating objects. Through a series of convolution and pooling layers, the resolution is reduced, the number of feature map channels is increased, and multi-scale waterlogged features are extracted.

[0055] Step 6: During upsampling in the decoder, the attention mechanism is used to calculate the weights associated with pixels and global features, highlighting water-logged pixels and suppressing the influence of floating objects and irrelevant pixels, thus guiding accurate segmentation.

[0056] Step 7: The decoder fuses feature maps of different scales, restores the resolution through convolution and upsampling, and finally outputs a binary image of the waterlogged area, where waterlogged pixels are 1 and non-waterlogged pixels are 0. The cross-entropy loss function is used to optimize the model for accurate identification of the waterlogged area. The cross-entropy loss function formula is:

[0057] Where N is the number of samples, y i is the true label, is the model prediction value;

[0058] Furthermore, the system is further provided with a water depth calculation unit, which is used to calculate the water depth, and the water depth calculation unit includes a dynamic water surface processing unit, which is used to calculate the water depth of the water area when there are floating objects on the dynamic water surface. The steps of calculating the water depth when there are floating objects on the dynamic water surface are as follows:

[0059] Step 1: Use the SIFT algorithm to extract stable feature points in adjacent frame images, avoiding interference from floating objects and locking the relatively stable area on the surface of the accumulated water;

[0060] Step 2: Combined with the optical flow method, assume that the pixel values ​​of the feature points of adjacent frame images are I(x, y, t) and I(x+Δx, y+Δy, t+Δt), solve the optical flow field according to the optical flow constraint equation, compensate for the dynamic water surface and calculate the depth after approximating the static state, where the optical flow constraint equation is I x u+I y v+I t =0, where I x , I y is the image partial derivative, I t is the time partial derivative, u and v are the pixel velocity components.

[0061] At the same time, a motion model is built based on the motion trajectory and speed of the floating object in the previous few frames to predict the occlusion effect of the current frame on the calculation of the water depth. The occluded area is dynamically adjusted or the suspicious area is marked for verification by multiple frames of data.

[0062] Furthermore, the water depth calculation unit also includes a slope correction unit, which is used to calculate the water depth when the water accumulation area is a slope and there are floating objects. The method of calculating the water depth when the water accumulation area is a slope and there are floating objects is as follows: using a known reference object in the image, setting the camera optical center O, the reference object projection point A, the water accumulation point P, the known actual height H of the reference object, and the pixel height h A , water point pixel height h P , according to the principle of similar triangles, the depth of water is

[0063] Taking into account the reflection and occlusion effects of floating objects, compare the water depth data around the reference object when there are no floating objects with the current frame data, and use statistical methods to correct the water depth in the affected area to ensure accurate depth calculation when there are floating objects on the slope.

[0064] It also includes a data transmission module, which is used to upload the water depth data and image information calculated by the image processing module to the cloud server or local monitoring center in real time. It uses 4G / 5G wireless communication technology to ensure smooth data interaction, ensure that the water depth data and image information can be uploaded in time, and the command information can be issued smoothly.

[0065] It also includes a self-learning optimization module. The self-learning optimization algorithm is started regularly to collect and acquire image data and calculate the water depth results. Under conditional calibration, it is compared with actual manual measurement or high-precision sensor data. If a large deviation is found, the deep learning model parameter weights are automatically adjusted, and the segmentation model of the waterlogged area strengthens the learning of pixel features of misjudged waterlogged boundaries. In the depth calculation module, according to the actual terrain, floating object distribution, and measurement error feedback, the optical flow method parameters, feature point tracking strategy, geometric correction model coefficients, and floating object motion prediction model parameters are optimized to improve the system accuracy and adaptability to cope with changes in waterlogging monitoring in different scenarios.

[0066] It also includes an auxiliary lighting system, which consists of an intelligent light sensor and low-power, high-brightness LED lights. It sets a preset threshold based on the ambient light intensity. When the light falls below the threshold (such as at night or in underground passages), the system automatically triggers the lighting to provide clear image acquisition conditions for the camera.

[0067] In addition, this device can also be configured with a solar power supply module. The module includes a solar panel. The solar panel is oriented in the appropriate lighting direction and connected to a rechargeable battery. It converts light energy into electrical energy for storage during the day and automatically powers the system at night or on cloudy days when there is insufficient light. This is particularly suitable for remote areas without mains electricity, reducing deployment and maintenance costs and ensuring stable operation of the system.

[0068] When using, first, the high-definition camera is installed in a suitable position to capture the road image at all times. When the light is insufficient, the auxiliary lighting system automatically lights up to ensure that the high-definition camera collects the cleaning image information. The collected image data is transmitted to the image processing module. The image processing module converts the color image into a grayscale image through the image preprocessing unit to reduce the amount of calculation, and then uses Gaussian filtering to erase small noise on the image. Then, in the face of complex floating object interference, use Mask first The R-CNN model finds the outline and position of floating objects, removes them from the image or marks them to obtain a relatively pure image. Then, the improved U-Net model downsamples to extract the water accumulation features while keeping a close eye on the water accumulation area through the attention mechanism, weakening irrelevant information. Finally, a black and white image of the water accumulation area is output to clearly divide the water accumulation range. With the water accumulation area, when calculating the depth, if a dynamic water surface is encountered, the SIFT algorithm is combined with the optical flow method to track the stable area of ​​water accumulation, calculate the water flow speed, "pull back" the dynamic water surface to a static state, and calculate the depth. For sloping terrain, the fixed reference objects in the image are used, based on the principle of similar triangles, combined with statistical methods to correct the depth deviation caused by floating objects and obtain the precise water accumulation depth.

[0069] In addition, the system can also compare the calculated water depth with the actual value measured manually through the self-learning optimization module, and adjust the model parameters if there is an error. Finally, the data transmission module reports the water depth and image information in a timely manner, providing key basis for urban flood control, traffic diversion and other decisions, and ensuring that the city safely overcomes the problem of waterlogging.

[0070] The basic principles, main features and advantages of the present invention are shown and described above. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention, and such changes and modifications fall within the scope of the invention as claimed.

Claims

1. A water depth monitoring and identification device based on artificial intelligence, characterized in that: include: High-definition camera, used to continuously collect road surface image data in the monitoring area; Image processing module, used to receive and process image data from high-definition cameras; The image processing module includes an image preprocessing unit and a waterlogging area recognition unit. The image preprocessing unit is used to perform grayscale and denoising processing on the image. The waterlogging area recognition unit is used to identify the waterlogging range. The steps of identifying the waterlogging range include: S1.1: Using the deep learning-based instance segmentation model Mask R-CNN, the preprocessed image is input. The model first uses a convolutional neural network to extract multi-scale feature maps of floating objects. S1.2: A region proposal network is used to slide a window on the feature map to generate anchor boxes of different scales and aspect ratios as candidate floating object regions. A small fully connected network is used to calculate the probability score of the anchor box containing the floating object and the offset relative to the actual floating object position. S1.3: For the candidate area after screening, determine whether it contains floating objects through the classification branch binary classification and output the classification probability P cls , the mask branch generates a pixel-level mask to outline the foreground candidate area, denoted as M; S1.4: According to the classification probability P cls Set a threshold, determine the floating object instances, and process the original image with the mask M to remove the floating objects or mark them as special areas, and output the purified image to the next step; S1.5: Receive the cleaned image and downsample it using a U-Net model encoder with an attention mechanism. After a series of convolution and pooling layers, the image resolution is reduced, the number of feature map channels is increased, and multi-scale water accumulation features are extracted. S1.6: When the decoder is upsampling, the attention mechanism intervenes to calculate the weights associated with pixels and global features, highlighting water-logged pixels and suppressing the influence of floating objects and irrelevant pixels. S1.7: The decoder fuses feature maps of different scales, restores the resolution through convolution and upsampling, and finally outputs a binary image of the waterlogged area. The cross-entropy loss function is used to optimize the model and accurately identify the waterlogged area. The cross-entropy loss function formula is: Where N is the number of samples, y i is the true label, is the model prediction value; It also includes a water depth calculation unit, which is used to calculate the water depth.

2. The artificial intelligence-based water depth monitoring and identification device according to claim 1, characterized in that: In step S1.5, the attention mechanism introduced by the U-Net architecture is used to enable the model to focus on key water features when segmenting the water area and ignore the interference of floating objects.

3. The artificial intelligence-based water depth monitoring and identification device according to claim 1 is characterized in that: The water depth calculation unit includes a dynamic water surface processing unit, which is used to calculate the water depth of the water area when there are floating objects on the dynamic water surface; The steps to calculate the water depth when there are floating objects on the dynamic water surface are: S2.1: Use the SIFT algorithm to extract stable feature points from adjacent frames, avoiding interference from floating objects and locking in relatively stable areas on the surface of the accumulated water; S2.2: Combined with the optical flow method, assume that the pixel values ​​of the feature points of adjacent frame images are I(x, y, t) and I(x+Δx, y+Δy, t+Δt), solve the optical flow field according to the optical flow constraint equation, compensate for the dynamic water surface and calculate the depth after approximating it to static, where the optical flow constraint equation is I x u+I y v+I t =0, where I x , I y is the image partial derivative, I t is the time partial derivative, u and v are the pixel velocity components.

4. The artificial intelligence-based water depth monitoring and identification device according to claim 3 is characterized by: While calculating the water depth when there are floating objects on the dynamic water surface, a motion model is built based on the motion trajectory and speed of the floating objects in the previous frames to predict the occlusion effect of the current frame on the calculation of the water depth. The occluded area is dynamically adjusted or suspicious areas are marked for verification by multiple frames of data.

5. The artificial intelligence-based water depth monitoring and identification device according to claim 3 is characterized in that: The water depth calculation unit further includes a slope correction unit, which is used to calculate the water depth when the water accumulation area is a slope and has floating objects; The method for calculating the water depth when the waterlogged area is a slope and there are floating objects is as follows: using the known reference object in the image, set the camera optical center O, the reference object projection point A, the waterlogged point P, the known actual height H of the reference object, and the pixel height h A , water point pixel height h P , according to the principle of similar triangles, the depth of water is 6. The artificial intelligence-based water depth monitoring and identification device according to claim 5, characterized in that: When calculating the water depth when the waterlogged area is a slope with floating objects, the water depth data around the reference object when there are no floating objects is compared with the current frame data based on the reflection and occlusion of the floating objects, and the water depth in the affected area is corrected using statistical methods.

7. The artificial intelligence-based water depth monitoring and identification device according to claim 1, characterized in that: It also includes a data transmission module, which is used to upload the water depth data and image information calculated by the image processing module to a cloud server or a local monitoring center in real time.

8. The artificial intelligence-based water depth monitoring and identification device according to claim 1, characterized in that: In step S1.1, the multi-scale features include but are not limited to the shape, texture, and color information of the floating objects.

9. The artificial intelligence-based water depth monitoring and identification device according to claim 1, characterized in that: It also includes a self-learning optimization module, which is used to regularly compare the collected images, the calculated water depth results and the actual manual measurement data. If there is a large deviation, the deep learning model parameter weights are automatically adjusted.

10. The artificial intelligence-based water depth monitoring and identification device according to claim 1, characterized in that: It also includes an auxiliary lighting system, which includes an intelligent light sensor for detecting light intensity and an LED lamp for adding a light source to the high-definition camera. A threshold is preset based on the ambient light intensity. When the light is dark below the threshold, the LED lamp is automatically triggered for lighting.

Images