CN117346846B - A method and device for automatically correcting water-measuring weir flow photography monitoring - Google Patents

Abstract

The invention relates to the technical field of hydrologic tests, in particular to an automatic-correction water weir flow photography monitoring method and device, comprising the steps of performing perspective correction on an original water gauge image, and performing weir groove inner wall segmentation processing by utilizing a SAM image segmentation model fused with a VIT model and a cross attention mechanism module to obtain a target weir groove inner wall segmentation result; obtaining a target weir groove water level according to a target weir groove inner wall segmentation result and a weir groove water level identification model; and obtaining the flow of the water measuring weir by using the water level flow relation model and the target weir groove water level. According to the invention, the SAM image segmentation model and the YOLO v5 target detection model are used for automatically correcting and identifying the weir groove water gauge image, so that the problems of high cost and limited stability of the conventional water gauge flow monitoring method are solved, and the method has good segmentation and identification capability on the water gauge or the inner wall of the weir groove under various environments or observation angles, improves the accuracy and efficiency of identification and monitoring, and has good universality.

Description

A method and device for automatically correcting water-measuring weir flow photography monitoring

Technical Field

The invention relates to the technical field of hydrological testing, and in particular to a method and a device for automatically correcting a water measuring weir flow photography monitoring method.

Background Art

A water measuring weir is a commonly used device for measuring water flow. Its working principle is to install weirs of specific size and geometry so that when the water flowing into the water measuring weir flows into the weir in a predetermined direction, the flow rate is calculated based on the water level difference between the weir and the weir. At present, the weir and the weir water level is usually monitored using float-type water level gauges or various ultrasonic open channel flow meters, radar water level gauges, etc. Among them, although float-type water level gauges are stable and reliable with good measurement accuracy, they usually require the construction of water level logging wells, which are costly. Ultrasonic flow meters and radar water level gauges are expensive and easily affected by water environment such as sediment, water temperature, and aquatic organisms.

In recent years, non-contact weir flow monitoring methods based on image processing have gradually attracted attention. For example, the patent with publication number CN115638835A discloses a weir flow automatic monitoring method and device based on image recognition algorithm. The method uses circular edge detection for the disk image of the float-type water level device, and performs opening operation and Hough line detection on the boundary line to obtain the connecting rod angle, thereby calculating the flow size; the patent with publication number CN113237534A discloses a rotating disk weir water level monitoring system. The monitoring and control module of the system builds YOLO v3 convolutional neural network, detects the number of isosceles triangles of the homemade water level gauge exposed above the water surface, and calculates the water level and flow of the weir according to the true height of the triangle. However, this method cannot accurately identify incomplete triangles exposed above the water surface; Hu Minquan et al. introduced a new type of water level gauge in "Automatic Measurement Technology of Water Head on Weir Based on Machine Vision", and proposed an automatic measurement algorithm in conjunction with the water level gauge. However, the recognition accuracy of the above-mentioned non-contact water-measuring weir flow monitoring method or device based on image processing is often limited by environmental conditions (light, shadow, aquatic plants, biological residues, etc.), and the generalization performance is limited; at the same time, error correction cannot be achieved. Once the marker is offset, the recognition accuracy is significantly reduced, the stability is limited, and the monitoring cost is expensive.

Summary of the invention

The present invention provides a method and device for automatically correcting a water-measuring weir flow photography monitoring method, which solves the technical problem that the existing water-measuring weir flow monitoring cost is expensive and the image water level recognition method has limited stability.

In order to solve the above technical problems, the present invention provides a method and device for automatically correcting flow photography monitoring of a water measuring weir.

In a first aspect, the present invention provides a method for automatically correcting a water measuring weir flow photography monitoring method, the method comprising the following steps:

Acquire an original water gauge image, and perform perspective correction on the original water gauge image to obtain a perspective corrected image;

Annotating foreground points and background points in the perspective-corrected image to generate annotated point information;

Obtain a pre-built SAM image segmentation model that integrates a VIT model and a cross-attention mechanism module, input the perspective-corrected image and the annotation point information into the SAM image segmentation model to perform weir inner wall segmentation processing, and obtain a target weir inner wall segmentation result; wherein the VIT model is used to extract image features of the perspective-corrected image; and the cross-attention mechanism module is used to decode and calculate the image features and the annotation point information;

Obtaining a target weir groove water level according to the target weir groove inner wall segmentation result and a pre-established weir groove water level recognition model;

The flow rate of the measuring weir is calculated using the water level-flow relationship model and the target weir water level.

In a further embodiment, before the step of obtaining the target weir water level according to the target weir inner wall segmentation result and the pre-established weir water level recognition model, the method further includes detecting the offset of the water gauge according to the target weir inner wall segmentation result, and when the water gauge is offset, correcting the marked point information using a pre-trained YOLO v5 target detection model, wherein the step of correcting the marked point information using the pre-trained YOLO v5 target detection model specifically includes:

The perspective-corrected image is identified using a pre-trained YOLO v5 target detection model to obtain the offset position coordinates of the water gauge;

Calculating the water gauge offset according to the water gauge offset position coordinates and the water gauge standard position coordinates;

The marking point information is corrected according to the water gauge offset to generate the corrected marking point information, and the original water gauge image is re-segmented according to the corrected marking point information through the SAM image segmentation model to dynamically update the target weir inner wall segmentation result.

In a further embodiment, the step of detecting the offset of the water gauge according to the target weir inner wall segmentation result comprises:

The boundary line between the measuring weir and the water gauge is determined according to the segmentation result of the inner wall of the target weir. The water gauge is detected for offset using the average angle between the boundary line between the measuring weir and the water gauge and the direction of the plumb line. If the average angle is not within the preset water gauge angle range, it is determined that the water gauge is offset.

In a further embodiment, the step of performing perspective correction on the original water gauge image to obtain the perspective corrected image further includes:

It is detected whether there is an area with missing pixel values in the perspective-corrected image. If so, the image is interpolated using the nearest neighbor interpolation method or the Lagrange interpolation method.

In a further embodiment, the step of inputting the perspective-corrected image and the annotation point information into the SAM image segmentation model to perform weir inner wall segmentation processing to obtain a target weir inner wall segmentation result comprises:

Extracting features of the perspective-corrected image using a VIT model to obtain image features;

Traversing the image features, decoding and calculating the image features and the annotation point information through a cross attention mechanism, and obtaining decoded image features and mask features;

The decoded image features are sequentially convolved, upsampled and processed by a multi-layer perceptron using the SAM image segmentation model to obtain the final image features. The mask features are dimensionally adjusted by a multi-layer perceptron to be consistent with the final image features. The two are multiplied to obtain the target weir inner wall segmentation result.

In a further embodiment, the step of obtaining the target weir water level according to the target weir inner wall segmentation result and the pre-established weir water level recognition model comprises:

Calculate the pixel area of the target weir inner wall by using the target weir inner wall segmentation result;

The pixel area of the target weir inner wall is input into a pre-established weir water level recognition model to obtain the target weir water level, wherein the weir water level recognition model is:

L＝h-k·S

Where L represents the target weir water level; h and k represent regression coefficients; and S represents the pixel area of the inner wall of the target weir.

In a further embodiment, the water level flow relationship model is:

Where Q represents the flow rate of the measuring weir; _Ce represents the flow rate empirical coefficient; θ represents the V-shaped weir top angle; g represents the gravitational acceleration; _he represents the height of the V-shaped weir top angle; and L represents the target weir water level.

In a second aspect, the present invention provides a flow photography monitoring device for a water measuring weir capable of automatic correction, the device comprising:

An image correction module is used to obtain an original water gauge image and perform perspective correction on the original water gauge image to obtain a perspective corrected image;

An image annotation module, used to annotate foreground points and background points in the perspective-corrected image to generate annotation point information;

An image segmentation module is used to obtain a pre-built SAM image segmentation model that integrates a VIT model and a cross-attention mechanism module, and input the perspective-corrected image and the annotation point information into the SAM image segmentation model to perform weir inner wall segmentation processing to obtain a target weir inner wall segmentation result; wherein the VIT model is used to extract image features of the perspective-corrected image; and the cross-attention mechanism module is used to decode and calculate the image features and the annotation point information;

A water level recognition module, used for obtaining a target weir groove water level according to the target weir groove inner wall segmentation result and a pre-established weir groove water level recognition model;

The flow monitoring module is used to calculate the water flow of the water measuring weir using the water level flow relationship model and the target weir water level.

The present invention provides a method and device for automatically correcting the flow photography monitoring of a water measuring weir. The method performs perspective correction on the original water gauge image, and uses the SAM image segmentation model that integrates the VIT model and the cross attention mechanism module to perform weir inner wall segmentation processing to obtain the target weir inner wall segmentation result; obtains the target weir water level according to the target weir inner wall segmentation result and the pre-established weir water level recognition model; and calculates the water measuring weir flow using the water level flow relationship model and the target weir water level. Compared with the prior art, the method uses the SAM model and the YOLO v5 target detection model to establish an automatically corrected weir water level image recognition model, and uses the water measuring weir water level and flow relationship model to obtain the real-time flow value. It has good recognition performance and adaptability for the weir water level and flow under various environments or observation angles, improves the accuracy and stability of the water measuring weir flow monitoring, is conducive to long-term continuous flow monitoring, and has the advantages of high monitoring efficiency, low cost, and wide application range.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic flow chart of a method for automatically correcting a water-measurement weir flow photography monitoring method provided by an embodiment of the present invention;

FIG2 is a schematic diagram of a flow monitoring device for a water measuring weir provided in an embodiment of the present invention;

FIG3 is a schematic diagram of a perspective-corrected image provided by an embodiment of the present invention;

4 is a schematic diagram of a perspective-corrected image with foreground points and background points marked according to an embodiment of the present invention;

FIG5 is a schematic diagram of a binary mask in a target weir inner wall segmentation result provided by an embodiment of the present invention;

6 is a schematic diagram of a weir image with prompt points and masks output by the SAM image segmentation model provided in an embodiment of the present invention;

FIG. 7 is a block diagram of an automatically correctable water-measurement weir flow photography monitoring device provided in an embodiment of the present invention.

DETAILED DESCRIPTION

The following specifically illustrates the implementation mode of the present invention in conjunction with the accompanying drawings. The embodiments are provided for illustrative purposes only and are not to be construed as limitations of the present invention. The accompanying drawings are provided for reference and illustration only and do not constitute limitations on the scope of patent protection of the present invention, because many changes may be made to the present invention without departing from the spirit and scope of the present invention.

Referring to FIG1 , an embodiment of the present invention provides a method for automatically correcting a water measuring weir flow photography monitoring method, which is applied to a water measuring weir flow monitoring device. As shown in FIG1 , the method includes the following steps:

S1. Obtain an original water gauge image, and perform perspective correction on the original water gauge image to obtain a perspective corrected image.

As shown in FIG2 , in this embodiment, a water gauge is installed on the inner wall of the channel or trough of the water measuring weir in a direction perpendicular to the water surface, and a column is fixed at a position at a suitable distance from the water gauge on one side of the water measuring weir. A time-delay monitoring camera and a solar power generation panel are mounted on the column. The solar power generation panel has a power of 100W and is equipped with a 60AH lithium battery. The solar power generation panel is used to power the monitoring camera. The camera preferably selects a camera model of Yutong YT-CAM8008LA-4G, with a resolution of 8 million pixels, a fixed focus of 4mm, a built-in 4G communication module, and a distance of about 100 meters from the ground. 3m, the water gauge is taken from above at about 60°, and the size of the picture is 3×3840×2160; it should be noted that in order to clearly see the water gauge scale from the monitoring screen and to be able to read the water level more accurately, this embodiment needs to adjust the camera lens angle so that the entire length of the water gauge is located in the camera monitoring screen and is as close to the center of the screen as possible. At the same time, this embodiment sets the time-delay camera shooting interval to 5 minutes. The camera takes water gauge pictures at regular intervals and uploads them to the cloud server simultaneously. The pictures are transmitted to the image processing module for water level calculation, and the monitoring personnel or spot check personnel can view and download water gauge pictures from the cloud.

Since the water gauge image taken from above will have perspective distortion, this means that in the image coordinate system, the real-world length of the water gauge corresponding to the unit pixel is inconsistent. Therefore, after the original water gauge image is captured by the surveillance camera, this embodiment also needs to perform perspective correction on the captured original water gauge image, correct it to the water gauge front view orientation, and obtain a perspective corrected image, wherein the calculation formula for perspective correction is:

Wherein, (x, y) represents the pixel coordinates of the original water gauge image; (m, n) represents the pixel coordinates of the perspective corrected image; M represents the spatial perspective transformation matrix; _a0 , _a1 , _a2 , _a3 represent linear transformation parameters; _b0 , _b1 represent translation parameters; _c0 , _c1 represent perspective transformation parameters.

In this embodiment, since the spatial perspective transformation matrix M has a total of 8 unknown quantities, at least 4 groups of corresponding points before and after correction are required to solve it. Therefore, this embodiment selects 4 points on the captured original water level gauge image, with coordinates of (1817.0, 966.4), (1856.0, 966.5), (1806.0, 1457.2), and (1841.4, 1457.6), which correspond to the intersections of the "9.5" and "0.5" scale lines of the water level gauge and the left and right edges of the water level gauge, respectively. The coordinates of the four points after correction are (450, 300), (550, 300), (450, 1600), and (550, 1600), respectively. In the image coordinate system, the straight lines formed by the two points on the left and right edges are both vertical lines. The perspective-corrected image obtained after spatial perspective correction is shown in Figure 3. In this embodiment, the size of the perspective-corrected image is 3×1000×2000.

It should be noted that after the original water level image is perspective corrected, if there are missing pixel values in some areas, the pixel value of the nearest non-missing point will be copied to the point or Lagrange interpolation will be used to obtain an image with missing value repaired, specifically: detect whether there is an area with missing pixel values in the perspective corrected image. If so, use the nearest neighbor interpolation method or Lagrange interpolation method to perform image interpolation.

S2. Marking foreground points and background points in the perspective-corrected image to generate marked point information.

This embodiment marks prompt points on the perspective-corrected image and assigns background indication information to the prompt points. The background indication information refers to whether the prompt point is a foreground prompt point (input is 1) or a background prompt point (input is 0). As shown in FIG4 , this embodiment preferentially marks three prompt points on the perspective-corrected image, wherein two prompt points are located on the inner wall of the weir groove with coordinates of (171, 896) and (880, 1103). These two prompt points are foreground points and are the areas of focus or interest of this embodiment; the third prompt point is located on the water ruler with coordinates of (475, 800). The third prompt point is a background point and is an irrelevant area of this embodiment.

S3. Obtain a pre-built SAM image segmentation model that integrates the VIT model and the cross-attention mechanism module, input the perspective-corrected image and the annotation point information into the SAM image segmentation model to perform weir inner wall segmentation processing to obtain the target weir inner wall segmentation result; wherein, the VIT model is used to extract the image features of the perspective-corrected image; and the cross-attention mechanism module is used to decode and calculate the image features and the annotation point information.

In this embodiment, the step of inputting the perspective-corrected image and the annotation point information into the SAM image segmentation model for segmentation processing to obtain the target weir inner wall segmentation result includes:

Extracting features of the perspective-corrected image using a VIT model to obtain image features;

Traversing the image features, decoding and calculating the image features and the annotation point information through a cross attention mechanism, and obtaining decoded image features and mask features;

The decoded image features are sequentially convolved, upsampled and processed by a multi-layer perceptron using the SAM image segmentation model to obtain the final image features. The mask features are dimensionally adjusted by a multi-layer perceptron to make them consistent with the final image features. The two are multiplied to obtain a target mask, that is, to obtain the target weir inner wall segmentation result.

Specifically, considering that the segmentation of the water gauge pixel area will interfere with the segmentation result due to the reflection of the water gauge, it is difficult to distinguish it from the water surface line of the water body by the naked eye. However, it is noted that the water surface line between the inner wall of the weir on which the water gauge relies and the water body is obviously white, which helps the model to identify the water surface line. For this reason, we use the SAM image segmentation model to perform mask segmentation on the inner wall. Among them, the SAM visual model integrates multiple text or visual classic models with Transformer modules, and performs image segmentation on the target area based on the prompt engineering. It encodes the image and annotation point information (annotation points, annotation boxes or text semantics) using the VIT model (Vision Transformer, visual self-attention model) and the Bert model respectively, and uses the cross-attention mechanism to calculate the pixel points or areas that are highly correlated with the annotation point information. It should be noted that the annotation point information needs to be continuously tried so that the SAM model can better understand and match the target area mask segmentation task.

The SAM image segmentation model traverses all image features obtained by the VIT model, and calculates whether each image feature is related to the annotation point information through the cross-attention mechanism. Among them, whether each image feature is related to the annotation point information is reflected by the weight matrix A. The weight matrix A obtained by softmax calculation can reflect the relative importance of each feature. The larger the value, the more relevant it is, and the smaller the value, the less relevant it is. If it is judged to be relevant, the corresponding pixel area outputs a mask "1", and if it is irrelevant, the mask "0" is output. The cross-attention mechanism is specifically as follows:

y＝Av

In the formula, A represents the weight matrix; q represents the image feature; k represents the annotation point information; k ^T is the transpose of k; D _h represents the feature dimension; v represents the value vector; y represents the decoded image feature.

The SAM image segmentation model performs convolution, upsampling and multi-layer perceptron processing on the image feature y output by the cross-attention mechanism module to obtain the final image feature. The mask feature output by the SAM model is adjusted in dimension by the multi-layer perceptron to make it consistent with the final image feature. The two are multiplied to obtain the target mask, and the mask with the highest confidence is retained through non-maximum suppression, and finally the mask matrix of the inner wall of the weir groove above the water surface with the highest confidence is output, that is, the target weir groove inner wall segmentation result. In the mask matrix (binarized matrix) of the inner wall of the weir groove above the water surface output by the SAM image segmentation model, the pixel area with a value of "1" is the target segmentation area, and the target weir groove inner wall segmentation result is shown in Figure 5. This embodiment can adjust the feature dimension by performing convolution, upsampling and multi-layer perceptron processing on the image feature output by the cross-attention mechanism module, which is convenient for calculation between tensors.

Considering the impact of water flow on the water gauge and the fatigue and aging of the fixing screws over time, the water gauge will deviate slightly. At this time, the marked points may not be able to play a good role in the SAM model. Therefore, before obtaining the target weir water level according to the target weir inner wall segmentation result and the pre-established weir water level recognition model, the water flow monitoring method provided by this embodiment also includes detecting the offset of the water gauge according to the target weir inner wall segmentation result, and when the water gauge deviates, the pre-trained YOLO v5 target detection model is used to correct the marked point information, specifically:

Determine the boundary line between the weir groove and the water gauge (the middle edge line of the segmentation result) according to the target weir groove inner wall segmentation result, calculate the average angle between the boundary line between the weir groove and the water gauge and the plumb line direction, and perform offset detection on the water gauge according to the average angle. If the average angle is not within the preset water gauge angle range, it is determined that the water gauge is offset and the marked point information needs to be corrected; if the average angle is within the preset water gauge angle range, it is determined that the water gauge is not offset.

When the water gauge is offset, the perspective-corrected image is identified using a pre-trained YOLO v5 target detection model to obtain the offset position coordinates of the water gauge;

Calculating the water gauge offset according to the water gauge offset position coordinates and the water gauge standard position coordinates;

The marking point information is corrected according to the water gauge offset to generate the corrected marking point information, and the original water gauge image is re-segmented according to the corrected marking point information through the SAM image segmentation model to dynamically update the target weir inner wall segmentation result.

Specifically, for ease of understanding, this embodiment provides an illustrative explanation of offset detection: observe the boundary line between the weir and the water gauge, calculate the average angle between the boundary line and the plumb bob direction, if the average angle is less than 3°, it is considered that the water gauge has not shifted; if the average angle exceeds 3°, it is considered that the water gauge has shifted, and the marking point information needs to be corrected.

The YOLO v5 target detection model in this embodiment adopts a network structure similar to Darknet, which includes multiple convolutional layers, pooling layers and fully connected layers, wherein the convolutional layer is used to extract image features, the pooling layer is used to reduce the dimension of the feature map, and the fully connected layer is used for final target detection. The loss functions used in the training process of the YOLO v5 target detection model include the prediction box regression loss function, the classification loss function, and the confidence loss function. The specific training process of the YOLO v5 target detection model is as follows:

The data set of perspective-corrected images is manually annotated, and the position of the water gauge in the image is annotated with a rectangular detection frame to form a training set consisting of about 100 samples. The training set includes water gauge images under different hydrological seasons (flood season, normal water season, and dry season) and environmental conditions (rainy days, foggy days, and nights), as well as coordinate information of the four vertices of the corresponding rectangular detection frame. In this embodiment, the training set is used to train a YOLO v5 target detection model to obtain a trained YOLO v5 target detection model for automatically identifying the position of the water gauge.

The trained YOLO v5 target detection model can better identify the position of the water ruler in the image through the rectangular detection frame and calculate the coordinates of the center point of the rectangular detection frame (x ₁ , y ₁ ); the offset of (x ₁ , y ₁ ) compared to the center coordinates of the water ruler rectangular frame (x ₂ , y ₂ ) before the offset is In this embodiment, the coordinates (m, n) of the marking point in the marking point information are corrected according to Δx and Δy, and the corrected coordinates of the marking point are obtained as (m', n'), and the corrected marking point information is obtained, wherein, The SAM image segmentation model re-executes the segmentation process according to the corrected annotation point information and dynamically corrects the mask of the inner wall of the target weir.

S4. Obtain the target weir water level according to the target weir inner wall segmentation result and the pre-established weir water level recognition model.

In this embodiment, the step of obtaining the target weir water level according to the target weir inner wall segmentation result and the pre-established weir water level recognition model comprises:

Calculate the pixel area of the target weir inner wall by using the target weir inner wall segmentation result;

The pixel area of the target weir inner wall is input into a pre-established weir water level recognition model to obtain the target weir water level.

Specifically, as shown in FIG6 , since there are a small number of pixel gaps at the edge of the inner wall mask of the weir (inner wall and water gauge, inner wall and water body), and the bottom of the weir is uneven, and there are cement blocks protruding on the inner wall surface, resulting in the water surface line in some areas not being level (the lower edge of the inner wall mask is not completely level), if the inner wall pixel height is directly calculated, a large error will be caused in water level recognition. However, by using the mask matrix to calculate the pixel area of the inner wall of the weir above the water surface line in the image coordinate system, the error can be reduced to a large extent. Therefore, in this embodiment, the water level is read through the water gauge scale line, and a linear relationship between the pixel area S of the inner wall of the weir and the true water level L is established to generate a weir water level recognition model, which is:

L＝h-k·S

Wherein, L represents the target weir water level; h and k represent regression coefficients; S represents the pixel area of the target weir inner wall. Since the mask matrix of the target weir inner wall segmentation result is a binary matrix, the pixel area of the target weir inner wall is the number of image pixels occupied by the inner wall. Therefore, in this embodiment, the target weir inner wall area can be obtained by summing the mask matrix of the target weir inner wall segmentation result. For example, when the water gauge range is 100 cm and the area S of the weir inner wall mask in the image coordinate system is 361826 pixels, the established weir water level recognition model is as follows:

L = 68.50-8.71×10 ^-5 ×S.

S5. Calculate the water flow rate of the water measuring weir using the water level flow relationship model and the target weir water level, wherein the water level flow relationship model is:

Where Q represents the flow rate of the measuring weir; _Ce represents the flow rate empirical coefficient; θ represents the V-shaped weir top angle; g represents the gravitational acceleration; _he represents the height of the V-shaped weir top angle; and L represents the target weir water level.

In this embodiment, the prompt point information correction and mask calculation are performed on the perspective corrected image, the target weir water level is calculated by the weir water level recognition model, and then the corresponding flow value is calculated by the water level-flow calculation formula of the measuring weir. The measuring weir in this embodiment is a V-shaped weir, and the coefficients and usage conditions of the water measuring weir flow calculation formula refer to the national standard JJG-711-1990.

The embodiment of the present invention provides a method for automatically correcting photographic monitoring of water measuring weir flow, the method comprising performing perspective correction on an original water gauge image, performing segmentation processing on the perspective corrected image by using a SAM image segmentation model that integrates a VIT model and a cross-attention mechanism module, establishing a relationship between mask pixels and the water level of a weir, and simultaneously using a YOLO v5 model to identify an offset distance of the water gauge caused by environmental factors, automatically correcting the segmentation mask, and having good segmentation and recognition capabilities for water gauges or inner walls of weirs in various environments or observation angles, thereby greatly improving the accuracy of water level and flow monitoring of water measuring weirs, and realizing rapid, continuous, low-cost, and automated monitoring of water level and flow of water measuring weirs, and having strong application value.

It should be noted that the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

In one embodiment, as shown in FIG. 7 , the present invention provides an automatically correctable water measuring weir flow photography monitoring device, the device comprising:

The image correction module 101 is used to obtain an original water gauge image and perform perspective correction on the original water gauge image to obtain a perspective corrected image;

An image annotation module 102 is used to annotate foreground points and background points in the perspective-corrected image to generate annotation point information;

The image segmentation module 103 is used to obtain a pre-built SAM image segmentation model that integrates the VIT model and the cross-attention mechanism module, input the perspective-corrected image and the annotation point information into the SAM image segmentation model for segmentation processing, and obtain the target weir inner wall segmentation result; wherein the VIT model is used to extract the image features of the perspective-corrected image; and the cross-attention mechanism module is used to decode and calculate the image features and the annotation point information;

The water level recognition module 104 is used to obtain the target weir water level according to the target weir inner wall segmentation result and the pre-established weir water level recognition model;

The flow monitoring module 105 is used to calculate the water measuring weir flow using the water level flow relationship model and the target weir water level.

For the specific limitations of an automatically correctable water-measuring weir flow photography monitoring device, please refer to the above-mentioned limitations of an automatically correctable water-measuring weir flow photography monitoring method, which will not be repeated here. A person of ordinary skill in the art will appreciate that the various modules and steps described in conjunction with the embodiments disclosed in this application can be implemented in hardware, software, or a combination of both. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

The embodiment of the present invention provides a flow photography monitoring device for a water measuring weir that can be automatically corrected. The device performs perspective correction on the original water gauge image through an image correction module; performs segmentation processing on the perspective corrected image through an image segmentation module; and realizes automatic monitoring of the water level and flow of the water measuring weir through a water level recognition module and a flow monitoring module. Compared with the prior art, the present application uses the YOLO v5 model to identify the offset distance of the water gauge caused by environmental factors, and automatically corrects the segmentation mask. The correction process is fast, real-time, stable and controllable, and effectively realizes the automatic segmentation and correction of the water gauge image, which has the advantages of simple implementation, high efficiency and low cost.

The above-mentioned embodiments only express several preferred implementation modes of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in the technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be based on the protection scope of the claims.

Claims (7)

1. A method for automatically correcting water flow measurement weir flow photography monitoring, characterized in that it includes the following steps: Acquire an original water gauge image, and perform perspective correction on the original water gauge image to obtain a perspective corrected image; Annotating foreground points and background points in the perspective-corrected image to generate annotated point information; Obtain a pre-built SAM image segmentation model that integrates a VIT model and a cross-attention mechanism module, input the perspective-corrected image and the annotation point information into the SAM image segmentation model to perform weir inner wall segmentation processing, and obtain a target weir inner wall segmentation result; wherein the VIT model is used to extract image features of the perspective-corrected image; and the cross-attention mechanism module is used to decode and calculate the image features and the annotation point information; Obtaining a target weir groove water level according to the target weir groove inner wall segmentation result and a pre-established weir groove water level recognition model; The flow rate of the water measuring weir is calculated by using the water level flow relationship model and the target weir water level. The step of inputting the perspective-corrected image and the annotation point information into the SAM image segmentation model to perform weir inner wall segmentation processing to obtain a target weir inner wall segmentation result comprises: Extracting features of the perspective-corrected image using a VIT model to obtain image features; Traversing the image features, decoding and calculating the image features and the annotation point information through a cross attention mechanism, and obtaining decoded image features and mask features; The decoded image features are sequentially convolved, upsampled and processed by a multi-layer perceptron using the SAM image segmentation model to obtain the final image features. The mask features are dimensionally adjusted by a multi-layer perceptron to be consistent with the final image features. The two are multiplied to obtain the target weir inner wall segmentation result. 2. A method for automatically correcting water-measurement weir flow photography monitoring according to claim 1, characterized in that, before the step of obtaining the target weir water level according to the target weir inner wall segmentation result and the pre-established weir water level recognition model, the method further comprises detecting the offset of the water gauge according to the target weir inner wall segmentation result, and when the water gauge is offset, correcting the marked point information using a pre-trained YOLO v5 target detection model, wherein the step of correcting the marked point information using the pre-trained YOLO v5 target detection model specifically comprises: The perspective-corrected image is identified using a pre-trained YOLO v5 target detection model to obtain the offset position coordinates of the water gauge; Calculating the water gauge offset according to the water gauge offset position coordinates and the water gauge standard position coordinates; The marking point information is corrected according to the water gauge offset to generate the corrected marking point information, and the original water gauge image is re-segmented according to the corrected marking point information through the SAM image segmentation model to dynamically update the target weir inner wall segmentation result. 3. The method for automatically correcting water-measurement weir flow photography monitoring according to claim 2, wherein the step of detecting the offset of the water gauge according to the segmentation result of the inner wall of the target weir groove comprises: The boundary line between the measuring weir and the water gauge is determined according to the segmentation result of the inner wall of the target weir. The water gauge is detected for offset using the average angle between the boundary line between the measuring weir and the water gauge and the direction of the plumb line. If the average angle is not within the preset water gauge angle range, it is determined that the water gauge is offset. 4. The method for automatically correcting the flow rate of a water measuring weir according to claim 1, characterized in that after the step of performing perspective correction on the original water gauge image to obtain the perspective corrected image, the method further comprises: It is detected whether there is an area with missing pixel values in the perspective-corrected image. If there is, the image is interpolated using the nearest neighbor interpolation method or the Lagrange interpolation method. 5. The method for automatically correcting water-measurement weir flow photography monitoring according to claim 1, characterized in that the step of obtaining the target weir channel water level according to the target weir channel inner wall segmentation result and the pre-established weir channel water level recognition model comprises: Calculate the pixel area of the target weir inner wall by using the target weir inner wall segmentation result; The pixel area of the target weir inner wall is input into a pre-established weir water level recognition model to obtain the target weir water level, wherein the weir water level recognition model is: L＝h-k·S Where L represents the target weir water level; h and k represent regression coefficients; and S represents the pixel area of the inner wall of the target weir. 6. The method for automatically correcting water-measurement weir flow photography monitoring according to claim 1, characterized in that the water level-flow relationship model is: Where Q represents the flow rate of the measuring weir; _Ce represents the flow rate empirical coefficient; θ represents the V-shaped weir top angle; g represents the gravitational acceleration; _he represents the height of the V-shaped weir top angle; and L represents the target weir water level. 7. A water-measuring weir flow photography monitoring device capable of automatic correction, characterized in that the device comprises: An image correction module is used to obtain an original water gauge image and perform perspective correction on the original water gauge image to obtain a perspective corrected image; An image annotation module, used to annotate foreground points and background points in the perspective-corrected image to generate annotation point information; An image segmentation module is used to obtain a pre-built SAM image segmentation model that integrates a VIT model and a cross-attention mechanism module, and input the perspective-corrected image and the annotation point information into the SAM image segmentation model to perform weir inner wall segmentation processing to obtain a target weir inner wall segmentation result; wherein the VIT model is used to extract image features of the perspective-corrected image; and the cross-attention mechanism module is used to decode and calculate the image features and the annotation point information; A water level recognition module, used for obtaining a target weir groove water level according to the target weir groove inner wall segmentation result and a pre-established weir groove water level recognition model; A flow monitoring module is used to calculate the flow of the water measuring weir using the water level flow relationship model and the target weir water level; The step of inputting the perspective-corrected image and the annotation point information into the SAM image segmentation model to perform weir inner wall segmentation processing to obtain a target weir inner wall segmentation result specifically includes: Extracting features of the perspective-corrected image using a VIT model to obtain image features; Traversing the image features, decoding and calculating the image features and the annotation point information through a cross attention mechanism, and obtaining decoded image features and mask features; The decoded image features are sequentially convolved, upsampled and processed by a multi-layer perceptron using the SAM image segmentation model to obtain the final image features. The mask features are dimensionally adjusted by the multi-layer perceptron to make them consistent with the final image features. The two are multiplied to obtain the target weir inner wall segmentation result.