Artificial-afforestation area extraction method based on high-resolution remote-sensing image

Abstract

Provided in the present invention is an artificial-afforestation area extraction method based on a high-resolution remote-sensing image. The method comprises: acquiring a remote-sensing image, and preprocessing the remote-sensing image; inputting the preprocessed remote-sensing image into an intelligent forest extraction model, so as to obtain a distribution map for all forests; performing edge detection and Hough transform line detection on the distribution map for all the forests, so as to obtain artificial afforestation regions and natural forest regions; and calculating the areas of the artificial afforestation regions and the areas of the natural forest regions, so as to obtain statistical results for artificial afforestation and natural forests. The present invention can solve the shortcomings of conventional monitoring means in terms of efficiency, cost and accuracy, and provides technical supports for large-scale forestry investigation and ecological monitoring.

Description

An artificial afforestation area extraction method based on high-resolution remote sensing images TECHNICAL FIELD

[0001] The present application belongs to the technical field of remote sensing image processing, image analysis and classification, and specifically relates to an artificial afforestation area extraction method based on high-resolution remote sensing images. BACKGROUND

[0002] Afforestation is an important measure to cope with global climate change and ecological environment protection. Monitoring the area and effect of afforestation is of great significance to achieve carbon sink targets, protect biodiversity, and develop scientific forestry policies. In particular, artificial afforestation, due to its clear arrangement rules, monitoring its coverage area can better evaluate the implementation effect and ecological benefit of afforestation projects.

[0003] Currently, the two most common ways to monitor artificial afforestation are supervised classification through remote sensing images to classify forest land, and manual delineation based on remote sensing images to determine the planting area of artificial afforestation through manual means.

[0004] Using supervised classification technology to distinguish forest land, the model has weak generalization ability for different regions and different vegetation types, and is prone to classification errors. At the same time, there are great problems in distinguishing artificial forests and natural forests, and it is difficult to accurately extract the distribution and area of artificial afforestation.

[0005] Using manual delineation, experts interpret remote sensing images visually and manually mark artificial afforestation areas to accurately extract the distribution and area of artificial afforestation, which has produced good results. However, using manual delineation also has the following shortcomings:

[0006] (1) Low efficiency: manual marking is slow and difficult to meet the monitoring needs of large areas.

[0007] (2) Strong subjectivity: differences in experience and judgment of different operators may lead to inconsistent results, affecting the objectivity of measurement.

[0008] (3) High cost: large-scale manual interpretation requires a large amount of manpower and financial investment, which is difficult to sustain in the long term.

[0009] SUMMARY

[0010] The present application aims to provide a technical method that utilizes high-resolution remote sensing images combined with machine learning and Hough transform algorithm, which can efficiently and accurately identify and measure the area of artificial afforestation, and effectively distinguish natural forest land and artificial forest land, improving the efficiency of monitoring, analysis and management.

[0011] The technical solution of the present application is as follows:

[0012] An artificial afforestation area extraction method based on high-resolution remote sensing images, the method comprising:

[0013] Obtaining a remote sensing image and pre-processing the remote sensing image;

[0014] Inputting the pre-processed remote sensing image into a forest land intelligent extraction model to obtain a distribution map of all forest lands;

[0015] Performing edge detection and Hough transform straight line detection on the distribution map of all forest lands to obtain an artificial afforestation area and a natural forest area;

[0016] Calculating the area of the artificial afforestation area and the natural forest area to obtain artificial afforestation and natural forest area statistical results.

[0017] Further, the forest land intelligent extraction model is constructed by a Swin transformer and an UperNet.

[0018] Further, the construction method of the forest land intelligent extraction model is:

[0019] Obtaining remote sensing image samples and forest land pixel-by-pixel delineation samples;

[0020] Preprocessing the obtained sample data, and dividing the preprocessed sample data into a training set and a data set;

[0021] Iteratively training the Swin transformer and the UperNet through the training set to obtain a forest land intelligent extraction model, and verifying the forest land intelligent extraction model through a verification set.

[0022] Further, the preprocessing is:

[0023] First, geometric correction is performed on the remote sensing image to remove image distortion caused by the sensor or satellite attitude, then radiation correction is performed to adjust the image brightness and contrast, and finally image enhancement is performed on the image, using contrast stretching and sharpening techniques to improve image quality and texture features.

[0024] Further, the edge detection and Hough transform straight line detection on the distribution map of all forest lands to obtain an artificial afforestation area and a natural forest area are specifically:

[0025] Performing gray scale processing on the extracted forest land area in the distribution map of all forest lands to convert the color image into a gray scale image;

[0026] Using a Canny edge detection algorithm to perform edge detection on the gray scale image to extract the significant edges in the image;

[0027] Set the accumulator threshold and line length threshold of the Hough transform algorithm, and detect the line in the image which meets the characteristics of artificial afforestation row and column;

[0028] According to the line distribution density and arrangement rule, the region with obvious row and column characteristics is identified as the artificial afforestation region, and the remaining region is the natural forest region.

[0029] Further, the artificial afforestation region and the natural forest region are calculated to obtain the artificial afforestation and natural forest area statistical result, which is specifically:

[0030] The pixel number of the artificial afforestation region and the natural forest region is counted respectively.

[0031] According to the resolution of the remote sensing image, the pixel number is converted into the actual area to obtain the artificial afforestation and natural forest area statistical result.

[0032] Further, the remote sensing image accuracy is greater than or equal to 0.5 meters.

[0033] Compared with the prior art, the present application has the following advantages:

[0034] The present application can solve the deficiencies of traditional monitoring methods in efficiency, cost and accuracy, and provide technical support for large-scale forestry investigation and ecological monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0035] The accompanying drawings generally illustrate various embodiments of the present application and, together with the description given, serve to explain the principles of the application. Wherever possible, the same reference numbers are used in the drawings and the description to refer to the same or similar parts. Such embodiments are illustrative, rather than restrictive, and are not intended to exhaustively enumerate all possible implementations and configurations.

[0036] Figure 1 shows a schematic diagram of the method steps of the present application;

[0037] Figure 2 shows a schematic diagram of the comparison between 0.5 meter and 1 meter remote sensing images of the present application;

[0038] Figure 3 shows a schematic diagram of the Hough transform principle of the present application;

[0039] Figure 4 shows a schematic diagram of the division structure of natural forest land and artificial afforestation land of the present application. DETAILED DESCRIPTION

[0040] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0041] The overall idea of the present application is: based on 0.5 meters or better than 0.5 meters of remote sensing image, the remote sensing image results of the research area are obtained after preprocessing. Then select a typical area to make a forest sample, train it, use the semantic segmentation method based on deep learning technology, realize the pixel-level recognition of forest on 0.5m image, get the forest recognition results, and finally use Hough transform algorithm for straight line detection, so as to distinguish the artificial afforestation area in the forest recognition results.

[0042] (I) remote sensing image

[0043] The precision of the remote sensing image used in the present application needs to reach or be better than 0.5 meters. A large number of experiments show that the remote sensing image with a precision lower than 0.5 meters cannot reflect the regularity that artificial afforestation is planted in rows and there is an obvious distance between rows, and the straight line detection effect of Hough transform algorithm is not good, which cannot achieve the expected effect.

[0044] (II) deep learning

[0045] Deep learning is a special machine learning that draws on the characteristics of the human brain composed of many neurons and completes thinking to form a model. It can be simply understood as a multi-layer neural network model, each layer uses the output of the previous layer as input, simulates the activity and interaction of neurons with mathematical models, and completes information processing

[0046] Semantic segmentation is an important direction in computer vision. Unlike object detection and recognition, semantic segmentation realizes pixel-level classification of images, which can be used for feature extraction and land use classification.

[0047] The present application adopts a deep learning-based semantic segmentation large model, uses Swin transformer as the backbone network, uses UperNet as the downstream semantic segmentation algorithm, and performs pixel-level segmentation and semantic assignment on the forest.

[0048] Swin transformer is a new type of Transformer architecture designed to address some limitations of traditional Transformer models when dealing with large-scale image data, such as fixed-size patch input and computational efficiency. It introduces hierarchical feature maps and a shifted window self-attention mechanism to achieve efficient processing of image data with linear computational complexity. Swin Transformer can combine deep image patches to generate hierarchical feature maps, adapting to different scales and high-resolution image processing. In addition, the shifted window mechanism improves model performance by connecting the windows of the previous layer, allowing the model to better capture local and global information. Swin Transformer has demonstrated state-of-the-art performance on multiple visual tasks, including COCO object detection and ADE20K semantic segmentation. Due to its efficiency and adaptability to different scales and resolutions, Swin Transformer has broad application prospects in computer vision and provides new ideas and possibilities for natural language processing and other fields. Unlike the low-resolution characteristic features of ViT (Vision Transformer) and the quadratic growth of complexity with image size, Swin Transformer is suitable as a backbone network for high-resolution input images and uses multiple times of downsampling results as features, making it suitable for the invention of extracting forest land from remote sensing images.

[0049] (Three) Hough transform algorithm, straight line detection

[0050] Hough transform is a feature extraction technique in image processing, which obtains a set of specific shapes by calculating the local maximum value of the cumulative result in a parameter space as the Hough transform result.

[0051] The Hough transform was first proposed by Paul Hough in 1962. The original Hough transform was designed to detect straight lines and curves. The original method required the analytic equation of the object boundary to be known, but did not require prior knowledge of the region's location. One of the outstanding advantages of this method is the robustness of the segmentation result, i.e. it is not very sensitive to incomplete data or noise. However, it is often impossible to obtain the analytic expression describing the boundary. The classical Hough transform was popularized by Richard Duda & Peter Hart in 1972. It was used to detect straight lines in images. Later, the Hough transform was extended to the recognition of objects of arbitrary shape, mostly circles and ellipses. The Hough transform uses a transformation between two coordinate spaces to map curves or lines of the same shape in one space to a peak at one point in the other coordinate space, thereby converting the problem of detecting arbitrary shapes into a statistical peak-finding problem.

[0052] The principle of the Hough transform for detecting straight lines is based on the idea of point-line duality. In the image space XY, a straight line can be represented by the equation y = px + q, where p is the slope and q is the intercept. However, the Hough transform converts this problem into the parameter space PQ, where p and q are the variables, while x and y are fixed values. In this way, a straight line in the image space corresponds to a point in the parameter space, and a point in the image space corresponds to a line in the parameter space.

[0053] Specifically, for each point (x, y) in the image space, a line q = y - px can be found in the parameter space. Each point (p, q) on this line represents a possible straight line passing through the point (x, y) in the image space. Therefore, points that are collinear in the image space will form multiple lines intersecting at the same point in the parameter space, and this intersection point corresponds to the straight line in the image space.

[0054] In practical applications, the straight line corresponding to the collinear points can be found by accumulating or voting in the parameter space. Specifically, a two-dimensional array A(p, q) can be created. For each point (x, y) in the image space, the corresponding line equation q = y - px is calculated, and the value at the corresponding position in the array A is incremented by 1. Finally, the peak points in the array A are found, which are the parameters of the straight lines corresponding to the collinear points in the image.

[0055] Artificial afforestation, trees have obvious row and column characteristics, so Hough transform algorithm can be used to detect straight lines and distinguish between natural forests and artificial afforestation.

[0056] Example 1

[0057] 1. Data acquisition and preprocessing

[0058] High-resolution remote sensing images are acquired using drones or satellites, ensuring a resolution of 0.5 meters or higher. The images are geometrically corrected to remove distortions caused by sensor or platform attitude, radiometrically corrected to adjust brightness and contrast, and image-enhanced using contrast stretching and sharpening techniques to improve image quality and detail performance.

[0059] 2. Forest extraction

[0060] Sample collection: Label representative forest samples to generate a training dataset

[0061] Model training: Train the model based on the sample set and evaluate its accuracy to obtain a forest intelligent extraction model

[0062] Forest intelligent extraction: Apply the trained model to high-resolution remote sensing images to obtain a distribution map of all forest areas, including artificial tree planting and natural forests.

[0063] 3. Straight line detection and artificial forest identification

[0064] Image preprocessing

[0065] Perform grayscale processing on the extracted forest area to convert color images to grayscale images.

[0066] Edge detection

[0067] Use the Canny edge detection algorithm to detect edges in the grayscale image and extract significant edges in the image.

[0068] Hough transform straight line detection

[0069] (1) Set the accumulator threshold and straight line length threshold of the Hough transform algorithm to detect straight lines in the image that meet the characteristics of artificial tree planting rows and columns.

[0070] (2) According to the distribution density and arrangement of straight lines, identify areas with obvious row and column characteristics as artificial tree planting areas.

[0071] 4. Area calculation

[0072] Region labeling

[0073] Label the identified artificial tree planting areas and natural forest areas to generate a classification result map.

[0074] Area conversion

[0075] (1) Pixel statistics: Count the number of pixels in different types of areas in the classified image.

[0076] (2) Area calculation: according to the resolution of remote sensing image, the number of pixels is converted into the actual area, and the statistical results of artificial afforestation and natural forest area are obtained.

[0077] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and inventive concept of the present application, can make equivalent replacements or changes within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for extracting artificial afforestation area based on high-resolution remote sensing images, characterized in that, The method comprises: acquiring remote sensing images and pre-processing the remote sensing images; inputting the pre-processed remote sensing images into a forest land intelligent extraction model to obtain a distribution map of all forest lands; performing edge detection and Hough transform straight line detection on the distribution map of all forest lands to obtain artificial afforestation areas and natural forest areas; performing area calculation on the artificial afforestation areas and the natural forest areas to obtain artificial afforestation and natural forest area statistical results. 2.The method for extracting artificial afforestation area based on high-resolution remote sensing image according to claim 1, characterized in that, The forest land intelligent extraction model is constructed by a Swin transformer and an UperNet. 3.The method for extracting artificial afforestation area based on high-resolution remote sensing image according to claim 1, characterized in that, The construction method of the forest land intelligent extraction model comprises: acquiring remote sensing image samples and forest land pixel-by-pixel delineation samples; pre-processing the obtained sample data and dividing the pre-processed sample data into a training set and a data set; iteratively training the Swin transformer and the UperNet through the training set to obtain a forest land intelligent extraction model, and verifying the forest land intelligent extraction model through a verification set. 4.The method for extracting artificial afforestation area based on high-resolution remote sensing image according to claim 1 or 3, characterized in that, The pre-processing comprises: firstly, performing geometric correction on the remote sensing images to remove image distortion caused by the sensor or satellite attitude, then performing radiation correction to adjust the image brightness and contrast, and finally performing image enhancement on the images, using contrast stretching and sharpening techniques to improve the image quality and texture features. 5.The method for extracting artificial afforestation area based on high-resolution remote sensing image according to claim 1, characterized in that, The edge detection and Hough transform straight line detection on the distribution map of all forest lands to obtain the artificial afforestation areas and the natural forest areas specifically comprises: performing grayscale processing on the extracted forest land areas in the distribution map of all forest lands to convert the color images into grayscale images; using a Canny edge detection algorithm to perform edge detection on the grayscale images to extract the significant edges in the images; setting the accumulator threshold and the straight line length threshold of the Hough transform algorithm to detect the straight lines in the images that meet the characteristics of artificial afforestation rows and columns; according to the straight line distribution density and arrangement rules, identifying the areas with obvious row and column characteristics as artificial afforestation areas, and the remaining areas as natural forest areas. 6.The method for extracting artificial afforestation area based on high-resolution remote sensing image according to claim 1, characterized in that, The area calculation on the artificial afforestation areas and the natural forest areas to obtain the artificial afforestation and natural forest area statistical results specifically comprises: respectively counting the pixel numbers of the artificial afforestation areas and the natural forest areas; according to the resolution of the remote sensing images, converting the pixel numbers into actual areas to obtain the artificial afforestation and natural forest area statistical results. 7.The method for extracting artificial afforestation area based on high-resolution remote sensing image according to claim 1, characterized in that, The accuracy of the remote sensing images is greater than or equal to 0.5 meters.

Images