Intelligent snow depth identification system and method

Abstract

This invention belongs to the field of resource and environmental technology, specifically relating to an intelligent snow depth recognition system and method. The system includes: a monitoring device and a processing system; the monitoring device is used to capture field images; the processing system is used to process the images captured by the monitoring device; the method includes: acquiring field images; calculating snow depth based on the acquired field images; automatically verifying and detecting anomalies in the calculated snow depth, and marking abnormal images; and identifying and updating the results of each marked abnormal image. This invention can adapt to various deformations of the benchmark, and the measured snow depth has a small error.

Description

Technical Field

[0001] This invention belongs to the field of resources and environment technology, and specifically relates to an intelligent snow depth recognition system and method. Background Technology

[0002] Snow depth refers to the vertical depth from the snow surface to the ground surface. Snow depth measurement data is essential basic data for applications such as regional snowmelt water resource assessment and watershed hydrological simulation. Currently, most field measurements of snow depth are conducted manually, using snow gauges or meter sticks perpendicular to the ground surface. To obtain long-term, continuous snow depth measurement data, surveyors create graduated markers and place them at fixed points, using a camera system to take periodic photos of the markers. The snow depth data is then obtained through automatic computer segmentation and interpretation of the photos.

[0003] Traditional snow depth image recognition based on computer graphics algorithms generally involves the following steps: 1) screening images of the area surrounding the marker covered with snow; 2) cutting out the marker shape using template matching; 3) determining the height of the marker not covered by snow in the matched template shape; and 4) determining the snow depth based on the original length of the marker and the identified height not covered by snow. This method has the following shortcomings in practical applications: 1) To ensure the camera can capture the entire marker, the camera's shooting angle is usually set relatively wide, resulting in images containing large areas of background. The complex and varied terrain and sky can affect the accuracy of the template matching method in cutting the marker image. 2) Outdoor weather conditions are unpredictable, with frequent strong winds. The wind causes varying degrees of swaying of the fixed camera, leading to distortions in the marker in the acquired images. In such cases, the marker height recognition result has a large error.

[0004] Therefore, there is an urgent need to invent a new snow depth identification system and method to improve the accuracy of snow depth measurement and identification. Summary of the Invention

[0005] This invention provides an intelligent snow depth recognition system, comprising: a monitoring device and a processing system;

[0006] The monitoring device is used to capture images in the field;

[0007] The processing system is used to process the images captured by the monitoring device.

[0008] Furthermore, the processing system includes:

[0009] The system includes a field image acquisition module, a snow image recognition module, a recognition result verification module, and an abnormal situation handling module.

[0010] The field image acquisition module is connected to the monitoring device and the snow image recognition module, and is used to acquire camera images at regular intervals and transmit the images to the snow image recognition module.

[0011] The snow image recognition module processes the images acquired by the field image acquisition module to calculate the snow depth;

[0012] The recognition result verification module automatically verifies the calculated snow depth and detects abnormal situations.

[0013] The abnormal situation handling module identifies and updates the results of images marked as abnormal one by one through human-computer interaction.

[0014] Furthermore, the monitoring device includes: a snow depth measuring rod and a timed camera;

[0015] The timed camera is aimed at the snow depth measuring pole to take a picture, so that the snow depth measuring pole falls within the imaging field of view of the timed camera;

[0016] Furthermore, the snow depth measuring pole consists of red and black blocks and numbers;

[0017] The red and black blocks are arranged alternately;

[0018] Each color block is 1 cm high;

[0019] A method for intelligent snow depth recognition includes:

[0020] Acquire field images;

[0021] Snow depth is automatically calculated based on acquired field images;

[0022] Automatically verify and detect anomalies in the calculated snow depth, and automatically annotate abnormal images;

[0023] Images marked as abnormal are identified one by one and the results are updated.

[0024] Furthermore, the specific method for calculating snow depth is as follows:

[0025] 1) Read the raw images acquired in the field;

[0026] 2) Select daytime images for further processing;

[0027] 3) Determine whether the ground is covered by snow based on the RGB histogram distribution of the image, and select images with snow cover for further processing;

[0028] 4) Use a pre-trained deep learning model to perform target detection and obtain the minimum bounding rectangle of the area where the benchmark is located;

[0029] 5) Trim the target area to be processed based on the minimum bounding rectangle;

[0030] 6) The red and black color blocks are segmented using the watershed algorithm from computer graphics;

[0031] 7) Use the contour line counting method to count the number of black blocks;

[0032] 8) Employ a pre-trained deep network model for digit recognition to identify digit systems exposed in images and locate the color patches closest to the snow surface;

[0033] 9) Count the number of pixels in the height dimension for complete and partially obscured color patches near the snow surface, and calculate the half-coverage.

[0034] The height of the masking block;

[0035] 10) Calculate the snow depth. The snow depth is equal to the height of the marker minus the cumulative height of the complete color block minus the height of the half-covered color block, and output the result.

[0036] Furthermore, the deep learning model in step 4) is a benchmark recognition network based on the improved U-Net architecture. The model input size is designed to be 864×864 pixels, and the output is the probability of each image pixel being identified as a benchmark.

[0037] Furthermore, the process of automatically verifying and detecting anomalies in the calculated snow depth is as follows:

[0038] 1) Automatically correct the results based on the constraint relationship between the red and black blocks in the benchmark;

[0039] 2) Draw time series curves and box plots based on continuous recognition results, and automatically label outliers.

[0040] Furthermore, the constraint relationships include:

[0041] 1) The number of black blocks is never more than the number of red blocks;

[0042] 2) The difference between the number of black blocks and the number of red blocks never exceeds 1.

[0043] Technical effects of the present invention:

[0044] The intelligent snow depth recognition system and method provided by this invention can adapt to various deformations of the markers, and the distortion of the markers does not affect the recognition of the system; the deep learning model is used to automatically detect the markers, so that the subsequent color block segmentation is not affected by the background color; the snow depth recognition result can be accurate to 0.1 cm, the maximum recognition error of a single image does not exceed 1 cm, the average error of the test at 4 detection points does not exceed 0.2 cm, and the proportion of abnormal detection results does not exceed 3%. Attached Figure Description

[0045] The accompanying drawings illustrate various embodiments generally by way of example rather than limitation, and are used, together with the specification and claims, to explain embodiments of the invention. Where appropriate, the same reference numerals are used in all drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the apparatus or method.

[0046] Figure 1 A diagram of the monitoring device of the present invention is shown;

[0047] Figure 2 A schematic diagram of the benchmark of the present invention is shown;

[0048] Figure 3 A schematic diagram of the processing system modules of the present invention is shown;

[0049] Figure 4 A schematic flowchart of the image processing system of the present invention is shown;

[0050] Figure 5 A schematic diagram of the snow depth intelligent recognition system of the present invention is shown. Detailed Implementation

[0051] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0052] This embodiment provides a snow depth intelligent recognition system and method;

[0053] The system includes a monitoring device and a processing system;

[0054] The snow depth field monitoring device consists of two core components: a snow depth measuring rod and a timed camera, such as... Figure 1 As shown, the snow depth measurement marker consists of red and black blocks and numbers, such as... Figure 2 As shown,

[0055] 1) Alternating bands of red and black color form a scale for measuring snow depth;

[0056] 2) Each color block is 1 cm high to ensure that the subsequent recognition error caused by image distortion does not exceed 1 cm;

[0057] 3) Numbers are marked for every two color blocks. Number recognition can help locate color blocks near the snow surface;

[0058] 4) Near-snow-covered areas are counted in image pixels along the height dimension to automatically calculate heights below 1 cm. A timed camera is positioned to capture images of the marker, ensuring the entire marker falls within the imaging field of view when there is no snow accumulation.

[0059] The processing system comprises four modules: a field image acquisition module, a snow image recognition module, a recognition result verification module, and an abnormal situation handling module. The processing system can adapt to various morphological characteristics of benchmarks in field images and is almost unaffected by special situations such as benchmark sway and imaging distortion.

[0060] The processing system provides a reliability probability value for the image recognition results;

[0061] The processing system supports human-machine interaction to verify the recognition results and correct abnormal results.

[0062] The field image acquisition module connects to the field monitoring device and acquires camera images at regular intervals.

[0063] The snow image recognition module calculates the snow depth based on the acquired image, and its processing flow is as follows: Figure 4 The process includes the following steps: 1) Reading the original images acquired in the field; 2) Reading the imaging time from the image metadata and selecting daytime images for subsequent processing; 3) Determining whether the ground is covered by snow based on the RGB histogram distribution of the images and selecting images with snow cover for subsequent processing; 4) Using a pre-trained deep learning model for target detection to obtain the minimum bounding rectangle of the area where the marker is located; 5) Cropping the target area to be processed based on the minimum bounding rectangle; 6) Using the watershed algorithm of computer graphics to segment red and black color blocks; 7) Using the contour counting method to count the number of black color blocks; 8) Using a pre-trained digit recognition deep network model to identify the digit system exposed in the image and locate the color block position closest to the snow surface; 9) Counting the number of pixels in the height dimension of complete color blocks and half-covered color blocks near the snow surface and calculating the height of the half-covered color blocks; 10) Calculating the snow depth, which is equal to the height of the marker minus the cumulative height of the complete color blocks minus the height of the half-covered color blocks, and outputting the result. The deep learning model in step 4) uses a dedicated benchmark recognition network based on an improved U-Net architecture. The model input size is designed to be 864×864 pixels, and the output is the probability that each image pixel is identified as a benchmark. Based on year-round images obtained from four field monitoring points on the Qinghai-Tibet Plateau, the model is trained using manually labeled high-quality samples, and the model with the smallest validation error is retained for intelligent recognition. During sample labeling, only the bounding rectangle of the benchmark is labeled.

[0064] The recognition result verification module automatically verifies the intelligent recognition results and detects anomalies based on standard values. Specifically, it: 1) automatically corrects the results based on the constraint relationship between the red and black blocks in the benchmark; 2) draws time series curves and box plots based on continuous recognition results and automatically labels outliers.

[0065] The constraints include: 1) the number of black blocks is never more than the number of red blocks; 2) the difference between the number of black blocks and the number of red blocks is never more than 1.

[0066] The abnormal situation handling module identifies and updates the results of images marked as abnormal through human-computer interaction.

[0067] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent recognition of snow depth, characterized in that, include: Acquire field images using monitoring devices; The monitoring device includes: a snow depth measuring rod and a timed camera; The timed camera is aimed at the snow depth measuring pole to capture an image, so that the snow depth measuring pole falls within the imaging field of view of the timed camera; the snow depth measuring pole is composed of red and black blocks and numbers, with the red and black blocks arranged alternately, and each block being 1 cm high; Snow depth is automatically calculated based on acquired field images; Automatically verify and detect anomalies in the calculated snow depth, and automatically annotate abnormal images; The images marked as abnormal are identified one by one and the results are updated accordingly. The specific method for calculating snow depth is as follows: 1) Read the raw images acquired in the field; 2) Select daytime images for further processing; 3) Determine whether the ground is covered by snow based on the RGB histogram distribution of the image, and select images with snow cover for further processing; 4) Use a pre-trained deep learning model for target detection to obtain the minimum bounding rectangle of the area where the snow depth measurement marker is located; 5) Trim the target area to be processed based on the minimum bounding rectangle; 6) The red and black color blocks are segmented using the watershed algorithm from computer graphics; 7) Use the contour line counting method to count the number of black blocks; 8) A pre-trained deep network model for digit recognition is used to identify digits exposed in the image and locate the color block position closest to the snow surface; 9) Count the number of pixels in the height dimension of complete and partially obscured color blocks near the snow surface, and calculate the height of the partially obscured color blocks; 10) Calculate the snow depth. The snow depth is equal to the height of the marker minus the cumulative height of the complete color block minus the height of the half-covered color block, and output the result. The process of automatically verifying and detecting anomalies in the calculated snow depth is as follows: 1) Automatically correct the results based on the constraint relationship between the red and black blocks in the benchmark; 2) Based on the continuous recognition results, time series curves and box plots are drawn, and outliers are automatically labeled; The constraints include: 1) The number of black blocks is never more than the number of red blocks; 2) The difference between the number of black blocks and the number of red blocks never exceeds 1. 2.The method of claim 1, wherein, The deep learning model in step 4) is a benchmark recognition network based on the improved U-Net architecture. The model input size is designed to be 864x864 pixels, and the output is the probability of each image pixel being identified as a benchmark.

Images