Method for automatically judging bare soil state change condition based on remote sensing image

Abstract

The invention belongs to the technical field of bare soil dynamic monitoring, and provides a bare soil state change condition automatic judgment method based on remote sensing images in order to solve the technical problems of large workload and low efficiency of current artificial explanation of bare soil state rechecking. Automatically classifying the state of each previous and later time phase bare soil monomer remote sensing image corresponding to the previous bare soil vector; secondly, a judgment rule of bare soil state change conditions is established, comparison is carried out based on state classification results of bare soil monomer remote sensing images of a front time phase and a rear time phase, and the state change type of each bare soil is obtained; and finally carrying out attribute assignment on the change type of the bare soil. Compared with manual visual interpretation, the method can achieve the quick rechecking of the bare soil state, improves the working efficiency, judges whether the bare soil is rectified or not and whether the rectification work is in place or not, and has a very high practical application value.

Description

Technical Field

[0001] The invention belongs to the technical field of bare soil dynamic monitoring, and in particular relates to a method for automatically determining bare soil state changes based on remote sensing images. Background Art

[0002] Dust refers to the particles formed by the dust on the surface being brought into the air by human or natural factors. It is the main source of particulate matter in my country's urban environment. How to quickly identify bare soil dust sources such as bare soil in construction, bare soil in demolition, bare ground, and material yards in cities with high precision, high frequency, and large scope, obtain the spatial location and area of ​​bare soil in a large area, and use multi-period images to continuously and dynamically monitor the changes in the state of bare soil, and obtain the rectification effect of bare soil patches is an effective means to solve the current low efficiency of environmental law enforcement and verification and the difficulty of precise control.

[0003] In recent years, remote sensing technology has been widely used due to its significant advantages of large-scale and periodic observations. Remote sensing images with high temporal and spatial resolution can provide reliable data support for bare soil monitoring. With the increasing maturity of deep learning remote sensing automatic extraction algorithms, automatic extraction of bare soil spatial location, area and other information in a large area has been achieved.

[0004] However, in the process of dynamic monitoring of bare soil, the current workflow still has significant shortcomings. In order to understand the rectification of bare soil in the past, it is necessary to manually compare the images before and after and verify the status of bare soil one by one. With the accumulation of monitoring cycles, the total amount of bare soil interpreted continues to increase, resulting in an exponential increase in the workload of review. The efficiency of manual comparison can no longer meet the needs of large-scale dynamic monitoring. Therefore, the development of an automatic review algorithm for bare soil status based on remote sensing images has become a key technical breakthrough to improve the effectiveness of environmental supervision. Summary of the invention

[0005] In order to solve the technical problems of large workload and low efficiency of current manual interpretation of bare soil status review, the present invention proposes an "automatic determination method of bare soil status change based on remote sensing images".

[0006] Automatic determination method of bare soil state change based on remote sensing images, such as Figure 1 As shown, the following steps are included:

[0007] Step 1: Data acquisition: obtain remote sensing images of the monitoring area before and after the monitoring area and the bare soil patch vector result files interpreted in the past;

[0008] Step 2, bare soil image clipping: clip the remote sensing images of the front and back phases of the monitoring area element by element according to the bare soil patch range of the bare soil patch vector result file to obtain the bare soil monomer remote sensing images of each front and back phase;

[0009] Step 3: Preparation of bare soil state classification dataset and model training.

[0010] Step 3.1, the bare soil status categories are divided according to the specific evolution of bare soil over time in the past. According to the proportion of the area covered by covering materials, the bare soil is divided into five bare soil status categories: no covering, low-proportion covering, medium-high proportion and full covering, covered with green plants, and formed into a building main body. Among them, no covering and low-proportion covering indicate that the bare soil has a large dust pollution problem and needs to be rectified; medium-high proportion and full covering, covered with green plants, and formed into a building main body indicate that the bare soil has been effectively rectified or there is no dust pollution problem; according to the five bare soil status categories, a bare soil status classification data set is constructed;

[0011] Step 3.2, dividing the bare soil state classification data set into a training set, a validation set, and a test set according to a preset ratio and performing training to obtain a bare soil state classification model;

[0012] Step 4: Use the trained bare soil state classification model to infer the bare soil single remote sensing image before and after each phase of step 2 to obtain the bare soil state category of each bare soil single remote sensing image before and after each phase;

[0013] Step 5: According to the bare soil status categories of each bare soil monomer remote sensing image before and after the step 4, a comparative analysis of the bare soil monomer remote sensing image status before and after the time phase is performed:

[0014] Automatically divided into three types of state changes: continuation, non-bare soil and recurrence.

[0015] Continuation: The status category of the single bare soil remote sensing image in the previous phase is no cover or low-proportion cover, and the status category of the single bare soil remote sensing image in the later phase is still no cover or low-proportion cover, indicating that the bare soil has not been effectively managed and is in a continuation state;

[0016] Non-bare soil: The status categories of the single-unit remote sensing images of bare soil in the previous phase are any of the following: no thatching, low-ratio thatching, medium-high ratio and full thatching, covered with green plants, and formed into the main building. The status categories of the single-unit remote sensing images of bare soil in the later phase are any of the following: medium-high ratio and full thatching, covered with green plants, and formed into the main building, indicating that the bare soil has been effectively managed and will not cause serious dust pollution.

[0017] Recurrence: The status categories of the single bare soil remote sensing images in the previous phase are medium-high ratio and full thatching, covered with green plants, and formed into a building body. The status categories of the single bare soil remote sensing images in the later phase are no thatching or low-ratio thatching, indicating that although the bare soil has been effectively treated in the early stage, recurrence has occurred, which will cause serious dust pollution;

[0018] Step 6. Assignment of bare soil vector state change type attributes: Create a new attribute field in all bare soil patch vector result files interpreted in the past, and assign the state change type result corresponding to each bare soil element to realize automatic determination of bare soil state change.

[0019] Technical effects:

[0020] The present invention first establishes a bare soil state classification model to automatically classify the states of each bare soil monomer remote sensing image of the previous and subsequent phases corresponding to the bare soil vector; secondly, a determination rule for the bare soil state change is established, and the state classification results of the bare soil monomer remote sensing images of the previous and subsequent phases are compared to obtain the state change type of each bare soil; finally, the bare soil change type is attributed. This method forms a complete set of bare soil dynamic continuous monitoring processes, and replaces the manual interpretation work by adopting an algorithm-automated classification method. Compared with manual visual interpretation, it can realize the rapid review of the bare soil state, improve work efficiency, reduce the workload of manual interpretation, and solve the current problem of large workload and low efficiency in bare soil state review. After verification in the project of dynamic and refined remote sensing monitoring of bare soil dust sources, the accuracy of the algorithm for automatic classification of bare soil status (no thatching, low-proportion thatching, medium-high-proportion and full thatching, covered with green plants, and formed into a building main body) of the bare soil monomer remote sensing images before and after the present invention can reach 85.08%, and the accuracy of automatic judgment of bare soil status changes (continuation, non-bare soil and recurrence) can reach 91.48%. Manual quality inspection is carried out on the basis of the above judgment results, which can save 60% of labor costs cumulatively compared with direct manual review, so that it can be judged whether the bare soil has been rectified and whether the rectification work is in place, which has high practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of the overall method of the present invention.

[0022] Figure 2 This is an example of a bare soil single remote sensing image of the previous and next phases obtained after clipping the bare soil patch vector surface element element by element. Among them: A-bare soil patch vector surface element; B-bare soil single remote sensing image of the previous phase; C-bare soil single remote sensing image of the next phase.

[0023] Figure 3 This is an example of remote sensing images of bare soil in various states. Among them: D-bare soil state category; E-remote sensing image example; M-no thatch; N-low-ratio thatch; X-medium-high-ratio and full thatch; Y-covered with green plants; Z-has formed the main body of the building. DETAILED DESCRIPTION

[0024] like Figure 1As shown, the present invention includes six steps, namely, data acquisition, bare soil image cropping, bare soil state classification model training, bare soil state automatic classification, bare soil state comparison analysis before and after time phases, and bare soil state change attribute assignment, so as to realize the automatic determination of bare soil state change.

[0025] Step 1, data acquisition: obtain the remote sensing images of the monitoring area before and after the time phase and the bare soil map vector result files interpreted in the past. The feature type of the bare soil map vector result file is the surface feature type.

[0026] Step 2: Bare soil image cropping: According to the bare soil patch range of the bare soil patch vector result file, crop the remote sensing images of the monitoring area before and after the time phases element by element to obtain the bare soil monomer remote sensing images of each before and after time phases, such as Figure 2 As shown in the figure, some example figures of bare soil monomer remote sensing images before and after the time phases obtained after clipping the remote sensing images before and after the time phases in step 1 element by element based on the bare soil map vector surface elements are listed.

[0027] Furthermore, the specific process of step 2 is as follows:

[0028] Step 2.1, load the previous bare soil spot vector result file in step 1 and the remote sensing image data before and after the monitoring area;

[0029] Step 2.2: Check the projection consistency to ensure that the bare soil spot vector result file has the same coordinate system as the remote sensing image data before and after the time phase. If they are different, projection conversion is required.

[0030] Step 2.3, traverse each bare soil spot vector surface element in the bare soil spot vector result file, and extract the geometric shape of each surface element;

[0031] Step 2.4: Crop the remote sensing images of the previous and next phases according to the geometric shape of the surface element to obtain the previous and next phases of bare soil monomer remote sensing images corresponding to the surface element.

[0032] Step 3: Preparation of bare soil state classification dataset and model training:

[0033] Step 3.1: Classify the bare soil status according to the specific evolution of bare soil over time. According to the proportion of bare soil covered by covering materials, it is divided into five bare soil status categories: no covering, low proportion covering, medium-high proportion and full covering, covered with green plants, and formed into the main building. Bare soil covering is to suppress the dispersion of surface soil particles under the action of wind through physical covering means (such as dustproof nets, geotextiles, etc.), thereby effectively reducing dust pollution. This measure can reduce the concentration of PM particles in the air.

[0034] At present, the commonly used covering material is dust net. By covering the bare soil with dust net, the bare soil status category is divided according to the area ratio of the dust net: no covering means that the dust net is not covered, and the ratio of the covered area to the total area of ​​the bare soil spot vector result file is 0%; low-ratio covering means that the ratio of the covered area to the total area of ​​the bare soil spot vector result file is less than or equal to 50%; medium-high ratio means that the ratio of the covered area to the total area of ​​the bare soil spot vector result file is greater than 50% and less than 100%, and full covering means that the ratio of the covered area to the total area of ​​the bare soil spot vector result file is 100%. Green plant coverage means that greening measures have been taken and no covering is required; the main building has been formed means that the building has been formed through engineering construction and no covering is required.

[0035] like Figure 3 As shown in the figure, there are examples of remote sensing images of bare soil in different states. Among them, the two categories of no covering and low-proportion covering indicate that the bare soil has a large dust pollution problem and needs to be rectified; the three categories of medium-high proportion and full covering, green plant coverage, and main building have been formed indicate that the bare soil has been effectively rectified or there is no dust pollution problem. A bare soil state classification dataset is constructed based on the five bare soil state categories.

[0036] Step 3.2: Divide the bare soil state classification dataset into training set, validation set and test set in a ratio of 7:2:1, and use the highly integrated YOLO11-cls classification method for training to obtain a bare soil state classification model.

[0037] Step 4: Use the trained bare soil state classification model to infer each bare soil monomer remote sensing image before and after the step 2 to obtain the bare soil state category of each bare soil monomer remote sensing image before and after the step 2.

[0038] Step 5: According to the bare soil state categories of each bare soil monomer remote sensing image before and after the step 4, the bare soil monomer remote sensing image states before and after the phase are compared and analyzed:

[0039] Automatically divided into three types of state changes: continuation, non-bare soil and recurrence.

[0040] Continuation: The status category of the single bare soil remote sensing image in the previous phase is no cover or low-proportion cover, and the status category of the single bare soil remote sensing image in the later phase is still no cover or low-proportion cover, indicating that the bare soil has not been effectively managed and is in a continuation state;

[0041] Non-bare soil: The status categories of the single-unit remote sensing images of bare soil in the previous phase are any of the following: no thatching, low-ratio thatching, medium-high ratio and full thatching, covered with green plants, and formed into the main building. The status categories of the single-unit remote sensing images of bare soil in the later phase are any of the following: medium-high ratio and full thatching, covered with green plants, and formed into the main building, indicating that the bare soil has been effectively managed and will not cause serious dust pollution.

[0042] Recurrence: The status categories of the single bare soil remote sensing images in the previous phase are medium-to-high proportion and full thatching, covered with green plants, and formed into a building main body. The status categories of the single bare soil remote sensing images in the later phase are no thatching or low-proportion thatching, indicating that although the bare soil has been effectively controlled in the early stage, recurrence has occurred, which will cause serious dust pollution.

[0043] Step 6: Assignment of bare soil vector state change type attributes: Create a new attribute field in all bare soil patch vector result files interpreted in the past, and assign the state change type result corresponding to each bare soil element to achieve automatic bare soil state change status.

[0044] The above implementation modes are only used to illustrate the present invention, but not to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the present invention. The patent protection scope of the present invention should be defined by the claims.

Claims

1. A method for automatically determining the change of bare soil state based on remote sensing images, characterized in that: The steps include: Step 1: Data acquisition: obtain remote sensing images of the monitoring area before and after the monitoring area and the bare soil patch vector result files interpreted in the past; Step 2, bare soil image clipping: clip the remote sensing images of the front and back phases of the monitoring area element by element according to the bare soil patch range of the bare soil patch vector result file to obtain the bare soil monomer remote sensing images of each front and back phase; Step 3: Preparation of bare soil state classification dataset and model training: Step 3.1, the bare soil status categories are divided according to the specific evolution of bare soil over time in the past. According to the proportion of the area covered by covering materials, the bare soil is divided into five bare soil status categories: no covering, low-proportion covering, medium-high proportion and full covering, covered with green plants, and formed into a building main body. Among them, no covering and low-proportion covering indicate that the bare soil has a large dust pollution problem and needs to be rectified; medium-high proportion and full covering, covered with green plants, and formed into a building main body indicate that the bare soil has been effectively rectified or there is no dust pollution problem; according to the five bare soil status categories, a bare soil status classification data set is constructed; Step 3.2, dividing the bare soil state classification data set into a training set, a validation set, and a test set according to a preset ratio and performing training to obtain a bare soil state classification model; Step 4: Use the trained bare soil state classification model to infer the bare soil single remote sensing image before and after each phase of step 2 to obtain the bare soil state category of each bare soil single remote sensing image before and after each phase; Step 5: According to the bare soil state categories of each bare soil monomer remote sensing image before and after the step 4, the bare soil monomer remote sensing image states before and after the phase are compared and analyzed: Automatically divided into three types of state changes: continuation, non-bare soil and recurrence. Continuation: The status category of the single bare soil remote sensing image in the previous phase is no cover or low-proportion cover, and the status category of the single bare soil remote sensing image in the later phase is still no cover or low-proportion cover, indicating that the bare soil has not been effectively managed and is in a continuation state; Non-bare soil: The status categories of the single-unit remote sensing images of bare soil in the previous phase are any of the following: no thatching, low-ratio thatching, medium-high ratio and full thatching, covered with green plants, and formed into the main building. The status categories of the single-unit remote sensing images of bare soil in the later phase are any of the following: medium-high ratio and full thatching, covered with green plants, and formed into the main building, indicating that the bare soil has been effectively managed and will not cause serious dust pollution. Recurrence: The status categories of the single bare soil remote sensing images in the previous phase are medium-high ratio and full thatching, covered with green plants, and formed into a building body. The status categories of the single bare soil remote sensing images in the later phase are no thatching or low-ratio thatching, indicating that although the bare soil has been effectively treated in the early stage, recurrence has occurred, which will cause serious dust pollution; Step 6, assigning attributes of bare soil status changes: Create a new attribute field in all bare soil patch vector result files interpreted in the past, and assign a value to the status change type result corresponding to each bare soil element to realize automatic determination of bare soil status changes.

2. The method for automatically determining the change of bare soil state based on remote sensing images according to claim 1 is characterized in that: The element type of the bare soil patch vector result file in step 1 is a surface element type.

3. The method for automatically determining the change of bare soil state based on remote sensing images according to claim 1 is characterized in that: The specific process of step 2 is as follows: Step 2.1, load the previous bare soil spot vector result file in step 1 and the remote sensing image data before and after the monitoring area; Step 2.2: Check the projection consistency to ensure that the bare soil spot vector result file has the same coordinate system as the remote sensing image data before and after the time phase. If they are different, projection conversion is required. Step 2.3, traverse each bare soil spot vector surface element in the bare soil spot vector result file, and extract the geometric shape of each surface element; Step 2.4: Crop the remote sensing images of the previous and next phases according to the geometric shape of the surface element to obtain the previous and next phases of bare soil single remote sensing images corresponding to the surface element.

4. The method for automatically determining the change of bare soil state based on remote sensing images according to claim 1 is characterized in that: In step 3.1, the covering material is a dust net, and no covering is defined as no dust net covering, and the ratio of the covered area to the total area of ​​the bare soil patch vector result file is 0%; Low-proportion thatching means that the ratio of the covered area to the total area of ​​the bare soil map vector result file is less than or equal to 50%; medium-high ratio means that the ratio of the covered area to the total area of ​​the bare soil map vector result file is greater than 50% and less than 100%; full thatching means that the ratio of the covered area to the total area of ​​the bare soil map vector result file is 100%; green plant coverage means that greening measures have been taken and no thatching is required; and the main building has been formed means that the building has been formed through engineering construction and no thatching is required.

5. The method for automatically determining the change of bare soil state based on remote sensing images according to claim 1 is characterized in that: In step 3.2, the bare soil state classification data set is divided into training set, validation set and test set in a ratio of 7:2:1, and the YOLO11-cls classification method is used for training to obtain the bare soil state classification model.

Images