A refined land use classification method for rare earth mining areas based on hyperspectral remote sensing imagery

Abstract

This invention discloses a refined land use classification method for rare earth mining areas based on hyperspectral remote sensing imagery, belonging to the field of remote sensing. Based on hyperspectral remote sensing imagery of rare earth mining areas, an object-oriented feature extraction module and an object-oriented one-dimensional convolutional neural network classification model based on a multi-layer attention mechanism are designed. These two modules complete the land use classification task for rare earth mining areas. By combining object-oriented features with convolutional neural networks, the classification accuracy for different land use categories is enhanced, while also avoiding "salt and pepper" noise in the classification results.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of remote sensing, and particularly relates to hyperspectral remote sensing, image processing and computer vision, mine monitoring and environmental protection. BACKGROUND

[0002] Land use survey is an important link in mine environmental monitoring. Through different classification algorithms, the land use result is obtained by classifying remote sensing images, which is an important means for land use survey of large-scale mine area. The development of remote sensing technology provides rich data sources for land use survey of mine area. High-resolution remote sensing images have great application potential in land use monitoring of mine area, but the cost is too high, and the image processing workload is also large. Rare earth mine points have the characteristics of small and scattered distribution. Although medium-resolution multispectral images such as Landsat can monitor a large area economically and labor-savingly, they cannot reflect the more detailed land use information of rare earth mine area. Hyperspectral remote sensing images not only have large image width and can be used for macro monitoring, but also have high spectral resolution, a large amount of information, and contain multiple narrow spectral bands, which are very sensitive to small spectral details and can obtain detailed land class spectral characteristics and differences, showing great potential in fine classification of land use.

[0003] With the continuous deepening of the research on automatic extraction of remote sensing information, the classification methods applied to land use information extraction in mine area are gradually enriched. Compared with pixel-based classification methods, object-oriented classification methods group image elements with the same characteristics into homogeneous objects as classification units, and the classification results can avoid the "salt and pepper" phenomenon. At the same time, object-oriented classification methods can fully consider the spectral characteristics and spatial characteristics of ground objects, and the overall effect of classification is better. However, due to the complex composition of surface features (mining working face, surface reclamation vegetation, etc.) in rare earth mine area, it is difficult to apply traditional object-oriented classification methods to fine classification of land use in rare earth mine area. With the development of artificial intelligence technology in image recognition and processing, the convolutional neural network algorithm with high efficient dimension reduction and high-level feature extraction ability has been initially applied to ground object classification and land use information extraction in open-pit mine area. Although the convolutional neural network classification model has achieved good classification accuracy, the classification results still cannot avoid the appearance of "salt and pepper" noise. Therefore, in the context of massive information of hyperspectral remote sensing data, how to better combine the advantages of object-oriented classification and convolutional neural network classification model to realize more efficient and more automated data analysis and scene interpretation is one of the most urgent problems to be solved. This will provide intelligent technical support for relevant regulatory departments to monitor the mining production and reclamation recovery of rare earth mine area. SUMMARY

[0004] Therefore, the present application is aimed at the ground object types and special environment of the mining area, combines object-oriented and convolutional neural network, and constructs an object-oriented-convolutional neural network rare earth mining area hyperspectral remote sensing image land use fine classification method, fully utilizes the advantages of hyperspectral remote sensing image data to improve the target recognition accuracy, speed and other performances of the model, can quickly and accurately identify each typical ground object target in the mining area, and obtains accurate land use classification information, thereby providing intelligent technical support for the relevant supervision departments to monitor the mining production and reclamation recovery of the rare earth mining area.

[0005] Therefore, the present application provides an implementation step of a rare earth mining area land use fine classification method for hyperspectral remote sensing images, which comprises the following steps.

[0006] Step 1: According to the current national standard and existing related research, combined with field survey information and the hyperspectral remote sensing image of the rare earth mining area, a rare earth mining area classification system is determined.

[0007] Step 2: An object-oriented feature extraction module is used, which is mainly responsible for extracting image features from the hyperspectral remote sensing image.

[0008] Step 3: An object-oriented one-dimensional convolutional neural network classification model based on a multi-layer attention mechanism (Object Convolutional on Transformer and CBAM, OCTC) (hereinafter referred to as the OCTC model) is constructed, deep image feature learning and classification of the hyperspectral remote sensing image are completed through the convolutional neural network, and then the rare earth mining area land use classification task is completed.

[0009] Preferably, the object-oriented feature extraction module has the following specific implementation steps: in the first step, according to the rare earth mining area classification system, an optimal segmentation scale of the hyperspectral remote sensing image is determined by using a scale parameter estimation model; in the second step, the optimal segmentation scale is applied to the segmentation of the hyperspectral remote sensing image to obtain the segmentation result of the hyperspectral remote sensing image; and in the third step, the image features of the hyperspectral remote sensing image are extracted from the segmentation result of the hyperspectral remote sensing image.

[0010] Preferably, the image features of the hyperspectral remote sensing image include spectral features and spatial features.

[0011] Preferably, the spectral features include band features and index features.

[0012] Preferably, the band features include 32 mean values and 32 standard deviations of each band.

[0013] Preferably, the index features are 32 original band reflectances of the hyperspectral remote sensing image as input parameters, and 8 index features are constructed, including chlorophyll vegetation index, normalized vegetation index, ratio vegetation coefficient, green normalized vegetation index, difference vegetation index, 2 normalized difference red edge indexes, and land chlorophyll index.

[0014] Preferably, the spatial features include geometric features and texture features.

[0015] Preferably, the texture features are constructed by the cooperativeness, contrast, dissimilarity, information entropy, mean, variance, correlation and second moment of the hyperspectral remote sensing image segmentation result based on the gray level co-occurrence matrix.

[0016] Preferably, the geometric features are calculated based on the hyperspectral remote sensing image segmentation result by introducing the idea of object-oriented classification, and the non-symmetry, boundary index, compactness, density, ellipse fitting, maximum closed ellipse radius, minimum closed ellipse radius, rectangular fitting, circularity, shape index, area, boundary length, length, aspect ratio, pixel number, volume and width of the hyperspectral remote sensing image are obtained.

[0017] Preferably, the OCTC model comprises the following steps: in the first step, a one-dimensional convolutional neural network is constructed to input the image features of the hyperspectral remote sensing image and extract deep image features thereof; the one-dimensional convolutional neural network comprises a convolutional layer, an activation function, a pooling layer and a full connection layer; in the second step, a Transformer module is introduced in the feature extraction process, which enables the model to dynamically focus on local image features and global image features, fully utilizes the global context, and thus improves the classification accuracy; in the third step, a convolutional attention module is used to extract significant feature information in the network, which is equivalent to an implicit image feature optimization, which is beneficial to mining the distinguishability in the rare earth mine land use classification task; and in the fourth step, the significant image feature information output by the one-dimensional convolutional neural network is classified to obtain land use classification information.

[0018] Compared with the prior art, the present application has the following advantages:

[0019] 1. The present application proposes a rare earth mine land use fine classification method for hyperspectral remote sensing image, and establishes a rare earth mine land use fine classification method framework for hyperspectral remote sensing image, which can obtain higher accuracy in rare earth mine ground object detection.

[0020] 2. This invention introduces an object-oriented classification approach, which can obtain richer spectral and spatial features to reduce interference from complex land features in rare earth mining areas and emphasize the feature differences between categories. At the same time, it solves the problem of noise in the classification results, and achieves the goal of extracting more accurate land use information in complex rare earth mining areas.

[0021] 3. This invention constructs an object-oriented one-dimensional convolutional neural network classification model based on a multi-layer attention mechanism. For a given object at a given scale, this model reshapes all its corresponding pixel-level features into a one-dimensional feature vector, then performs deep feature learning and prediction, and finally completes the rare earth mining area land use classification task. Attached Figure Description

[0022] Figure 1 A framework diagram of the implementation steps of the method for refined land use classification of rare earth mining areas based on hyperspectral remote sensing images provided by the present invention;

[0023] Figure 2 This invention provides a comparison of the classification results of the classification method with other classification methods. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will be further described below with reference to embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without creative effort are within the scope of protection of the invention.

[0025] This embodiment is based on hyperspectral remote sensing images of the Lingbei rare earth mining area in southern Jiangxi.

[0026] A method for refined land use classification in rare earth mining areas based on hyperspectral remote sensing imagery is provided, and the implementation steps are outlined in the diagram below. Figure 1 As shown, the implementation steps of this method include:

[0027] Step 1: Based on current national standards and relevant existing research, combined with field survey information and hyperspectral remote sensing images of the Lingbei rare earth mining area, determine the classification system for rare earth mining areas;

[0028] Step 2: Utilize the object-oriented feature extraction module, which is primarily responsible for extracting image features from hyperspectral remote sensing images; such as... Figure 1 The hyperspectral image in the object-oriented feature extraction module shown is the one obtained in step 1;

[0029] Step 3: Construct the OCTC model, and use a convolutional neural network to complete the deep image feature learning and hyperspectral remote sensing image classification, thereby completing the land use classification task for rare earth mining areas.

[0030] Preferably, the hyperspectral remote sensing image feature extraction module, such as Figure 1 As shown in the (object-oriented feature extraction module), its specific implementation steps are as follows: First, based on the rare earth mining area classification system, the optimal segmentation scale of the hyperspectral remote sensing image is determined using a scale parameter estimation model; Second, the optimal segmentation scale is applied to the segmentation of the hyperspectral remote sensing image to obtain the hyperspectral remote sensing image segmentation result; Third, the image features of the hyperspectral remote sensing image are extracted from the hyperspectral remote sensing image segmentation result.

[0031] Preferably, the image features of the hyperspectral remote sensing image include spectral features and spatial features.

[0032] Preferably, the spectral features include band features and exponential features; the spatial features include geometric features and texture features.

[0033] Preferably, the spectral features include 32 means and 32 standard deviations for each band.

[0034] Preferably, the index features are constructed using the reflectance of 32 original bands of the hyperspectral remote sensing image as input parameters. The index features include chlorophyll vegetation index, normalized vegetation index, ratio vegetation coefficient, green normalized vegetation index, difference vegetation index, two normalized difference red edge indices, and terrestrial chlorophyll index.

[0035] Preferably, the texture features are constructed from the synergy, contrast, dissimilarity, information entropy, mean, variance, correlation, and second moment of the hyperspectral remote sensing image segmentation results based on the gray-level co-occurrence matrix.

[0036] Preferably, the geometric features are derived from the concept of object-oriented classification. Based on the segmentation results of hyperspectral remote sensing images, the following parameters are calculated: asymmetry, boundary index, compactness, density, ellipse fitting, radius of the largest closed ellipse, radius of the smallest closed ellipse, rectangle fitting, roundness, shape index, area, boundary length, length, aspect ratio, number of pixels, volume, and width of the hyperspectral remote sensing image.

[0037] The specific information regarding the image features is shown in Table 1:

[0038] Table 1. Specific Information Regarding Features

[0039]

[0040] The OCTC model, such asFigure 1 As shown in the OCTC model, its specific implementation steps are as follows: First, a one-dimensional convolutional neural network is constructed to extract deep features from the original input features; the one-dimensional convolutional neural network includes convolutional layers, activation functions, pooling layers, and fully connected layers; Second, a Transformer module is introduced in the feature extraction process, which allows the model to dynamically focus on local and global features, making full use of the global context, thereby improving classification accuracy; Third, a convolutional attention module is used to extract salient feature information in the network, which is equivalent to an implicit feature optimization, which is beneficial for mining discriminability in classification tasks; Fourth, the salient feature information output by the one-dimensional convolutional neural network is classified to obtain land use classification information.

[0041] Compared with existing technologies, this invention does not require cumbersome operations and also exhibits excellent performance in classification efficiency and accuracy. Land use type classification was performed on the original images of this embodiment using two typical machine learning methods, a one-dimensional convolutional neural network classification method, and the classification method of this invention. The classification accuracy is shown in Table 2. The typical machine learning methods are K-nearest neighbor classification and random forest classification. The classification results are as follows: Figure 2 As shown, Figure 2 (a) is the original image of the rare earth mining area in the Lingbei mining area. Figure 2 (b) shows the results of the K-nearest neighbor classification method. Figure 2 (c) is a graph showing the results of the random forest classification method. Figure 2 (d) shows the classification results of the one-dimensional convolutional neural network method. Figure 2(e) shows the classification results of the present invention. Comparing the classification results with the rare earth mining area's land cover categories, it can be seen that, based on object-oriented principles, all classification methods effectively avoid the "salt and pepper" phenomenon. However, among the classification results of machine learning algorithms, the K-nearest neighbor classification method has the worst classification performance. A comparison of the classification results with the original image shows that the K-nearest neighbor classification method struggles to accurately distinguish between cultivated land and orchards, and also misses many classifications of transportation land. While the random forest classification method generally performs well, misclassification is common in some details, especially in areas with multiple land cover couplings. For example, in areas coupling mining areas and reclaimed vegetation, the extraction of reclaimed vegetation is poor. Furthermore, there is significant confusion between reclaimed vegetation and orchards. This is because the classification features of reclaimed vegetation and orchards are quite similar, with only minor differences, making the random forest classification method less effective at distinguishing between them. The classification method of the present invention exhibits the best classification effect, not only with the highest accuracy, but also, as can be seen from the results, it has a greater advantage than traditional machine learning algorithms in handling areas with fragmented and small land features, significantly reducing misclassification and omissions, and effectively improving the extraction accuracy of fragmented land features. For land features with similar classification characteristics, the classification method of the present invention also has higher accuracy, effectively distinguishing between reclaimed vegetation and orchards. In rare earth mining area environmental monitoring tasks, the classification method of the present invention can accurately and efficiently identify typical land features within the mining area, providing excellent technical means and support for the information-based supervision of rare earth mining areas.

[0042] Table 2 Comparison of classification accuracy of different methods

[0043]

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and scope of the present invention should be included within the protection scope of the present invention.

Claims

1. A rare earth mine area land use refinement classification method for hyperspectral remote sensing images, characterized by: The implementation steps of the method are: Step 1: combining field survey information and hyperspectral remote sensing images of the rare earth mine area to determine a rare earth mine area classification system; Step 2: using an object-oriented feature extraction module to extract image features from the hyperspectral remote sensing images; Step 3: constructing an object-oriented one-dimensional convolutional neural network classification model based on a multi-layer attention mechanism, completing deep feature learning and classification of the hyperspectral remote sensing images through the convolutional neural network, and then completing the rare earth mine area land use classification task; The object-oriented feature extraction module has the following specific implementation steps: first, using a scale parameter estimation model to determine the optimal segmentation scale of the hyperspectral remote sensing images according to the rare earth mine area classification system; second, applying the optimal segmentation scale to the segmentation of the hyperspectral remote sensing images to obtain the hyperspectral remote sensing image segmentation results; third, extracting the image features of the hyperspectral remote sensing images from the hyperspectral remote sensing image segmentation results; The image features of the hyperspectral remote sensing images include spectral features and spatial features; The spectral features include band features and index features; The index features are constructed by taking 32 original band reflectances of the hyperspectral remote sensing images as input parameters, and the index features are chlorophyll vegetation index, normalized vegetation index, ratio vegetation coefficient, green normalized vegetation index, difference vegetation index, two normalized difference red edge indices, and land chlorophyll index; The object-oriented one-dimensional convolutional neural network classification model based on a multi-layer attention mechanism has the following specific implementation steps: first, constructing a one-dimensional convolutional neural network and inputting the image features of the hyperspectral remote sensing images to extract deep image features; the one-dimensional convolutional neural network includes convolutional layers, activation functions, pooling layers, and fully connected layers; second, introducing a Transformer module into the one-dimensional convolutional neural network to dynamically focus on local image features and global image features; third, using a convolutional attention module to extract significant image feature information in the one-dimensional convolutional neural network, optimizing the image features, and mining the distinguishability of the land use classification task; fourth, classifying the significant image feature information output by the one-dimensional convolutional neural network to obtain land use classification information.

2. The rare earth mine land use refined classification method for hyperspectral remote sensing images of claim 1, wherein: The band features are 32 mean values and 32 standard deviations of each band.

3. The method for hyperspectral remote sensing image-oriented refined classification of land use in rare earth mining area according to claim 1, characterized in that: The spatial features include geometric features and texture features.

4. The rare earth mine land use refined classification method for hyperspectral remote sensing images of claim 3, wherein: The texture features are constructed by the cooperativeness, contrast, dissimilarity, information entropy, mean value, variance, correlation, and second moment of the hyperspectral remote sensing image segmentation results based on a gray level co-occurrence matrix.

5. The method for hyperspectral remote sensing image-oriented refined classification of land use in rare earth mining area according to claim 3, characterized in that: The geometric features are calculated based on the hyperspectral remote sensing image segmentation results by introducing the idea of object-oriented classification, and include asymmetry, boundary index, compactness, density, ellipse fitting, radius of the maximum closed ellipse, radius of the minimum closed ellipse, rectangular fitting, circularity, shape index, area, boundary length, length, aspect ratio, pixel number, volume, and width.

Images