Positive sample data set construction method for astronomical image denoising application

Abstract

The invention discloses a positive sample data set construction method for astronomical image denoising application, relates to the field of image processing, and solves the problems that data sets for denoising and stray light removing tasks of astronomical images are relatively deficient and the processing performance needs to be improved at present. The method comprises the following steps: acquiring and cutting star maps at different time periods; identifying and positioning star points; generating and matching a double-triangle structure; and star map alignment and data set positive and negative sample selection are realized. The method is suitable for model training data construction of astronomical image denoising and stray light removal processing. In the subgraph cutting process, an unpolluted or lightly polluted area is preferentially reserved, the effectiveness and pertinence of data are improved, a double-triangle structure is constructed by accurately recognizing star points subsequently, star point matching and image alignment are carried out by utilizing a geometric hash algorithm, and finally, proper images are selected as positive and negative samples of a data set. Under the same astronomical observation condition, the method is better in denoising effect and higher in data processing efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of image denoising, and in particular to a method for constructing a positive sample data set for astronomical image denoising applications. Background Art

[0002] Denoising and removing stray light from astronomical images is an important task in the field of computer vision and is widely used in celestial body recognition, astronomical data processing, observation accuracy improvement, and space exploration. Existing datasets perform well in the case of weak stray light interference, but they still face great challenges for astronomical images with strong stray light and complex noise.

[0003] The relative scarcity of datasets for astronomical image denoising tasks limits the training and performance improvement of existing algorithms. The acquisition process of astronomical images is limited by the observation equipment, observation conditions and the optical properties of celestial bodies, and it is difficult to obtain a large amount of high-quality, noise-free real data at the same time. In order to solve this problem, some existing studies usually use artificial simulation of noise or stray light sources on clean star maps to generate training data. Although these simulated data can help the network learn denoising features to a certain extent, due to the complexity and diversity of noise types in real astronomical images, including photoelectric noise, atmospheric disturbances, thermal noise generated by equipment, and interference from stray light, these synthetic data are difficult to fully cover all situations in the real observation environment. Therefore, although the existing datasets can play a certain role in training denoising models, it is difficult to fully meet the needs of real astronomical scenes, resulting in the model's performance in practical applications is not ideal, especially when dealing with complex weak-signal celestial bodies and strong stray light interference, the denoising effect still has a lot of room for improvement.

[0004] The present invention proposes a method for constructing a data set using real images taken by astronomical telescopes. The star images taken come from multiple observations of the same observation area at different time periods, so that the noise and stray light in the data set are closer to the real observation environment. This method not only reduces the complexity of data acquisition, but also effectively improves the performance of the denoising model when processing actual astronomical images. Summary of the invention

[0005] In order to solve the problems in the prior art that there is a relative shortage of data sets for astronomical images for denoising and stray light removal tasks and that the processing performance needs to be improved, the present invention provides a method for constructing a positive sample data set for astronomical image denoising applications. The method of the present invention is mainly suitable for constructing model training data for astronomical image denoising and stray light removal. In the process of sub-image cropping, the present invention gives priority to retaining unpolluted or lightly polluted areas, thereby improving the validity and pertinence of the data, and subsequently accurately identifies and locates star points through a local maximum detection algorithm, constructs a double triangle structure, and uses a geometric hashing algorithm to perform star point matching and image alignment, and finally selects appropriate images as positive and negative samples of the data set. Under the same astronomical observation conditions, the data set constructed by the present invention can enable the model to have a better denoising effect and higher data processing efficiency.

[0006] The method for constructing a positive sample dataset for astronomical image denoising applications includes the following steps:

[0007] Step 1: obtaining star maps of different time periods, selecting the same area of ​​the star maps for cropping, and obtaining two sub-images;

[0008] Step 2: Star point identification and positioning;

[0009] For each cropped sub-image, star point identification and positioning are performed through local maximum detection;

[0010] Step 3: Generation and matching of double triangle structures;

[0011] The star points identified in the two sub-images are represented by a double triangle structure to represent the geometric relationship between the star points; the corresponding star points in the two sub-images are matched;

[0012] Step 4: According to the result of star point matching in step 3, complete the star map alignment;

[0013] Step 5: Selection of positive and negative samples of the data set;

[0014] According to the results of star map alignment in step 4, the unpolluted areas and strongly polluted areas in the star map are cropped to generate a data set.

[0015] Beneficial effects of the present invention:

[0016] The present invention starts with star point recognition and geometric relationship modeling, and processes the sub-image area of ​​the astronomical image for subsequent star map alignment and data set construction. The present invention crops the star maps taken at different times, uses the local maximum detection algorithm to accurately identify and locate the star points in the sub-image, and improves the accuracy of star point detection through a dynamic adaptive threshold method. After the star point is identified, a double triangle structure is constructed according to the recognition result, and the geometric relationship features between the star points, including the side length and angle, are extracted by selecting four star points and constructing two non-intersecting triangles, and a geometric hash code is generated to achieve efficient star point matching. After the star point matching is completed, the coordinates of the star points in the two star maps are uniformly transformed using affine transformation to ensure that the matched star points can be accurately aligned in the original coordinate system, thereby improving the accuracy of star map alignment. After the star map alignment is completed, the more seriously polluted areas and the cleaner areas are cropped according to the degree of light pollution, so as to generate a diversified training data set for the astronomical image denoising model.

[0017] The present invention solves the problems of low star point matching accuracy and insufficient generalization ability of denoising models under light pollution and noise conditions by cropping star maps based on real scenes, accurately identifying star points, modeling geometric relationships and affine alignment, and adopting cropping strategies for different regions. Ultimately, the constructed denoising dataset can provide higher denoising performance and better star point retention effects in the actual application of the model, achieving a significant improvement in the efficiency and quality of astronomical image processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The present invention is a flowchart of a method for constructing a positive sample data set for astronomical image denoising applications.

[0019] Figure 2 This is a flow chart of the double triangle structure matching method.

[0020] Figure 3 The overall schematic diagram of the double triangle structure matching star map. Among them, (a) represents the strong and weak contaminated star map, (b) represents the image after the local maximum detection algorithm detects the star point, c(a) represents the schematic diagram of the double triangle structure, and (a) represents the image aligned after affine transformation.

[0021] Figure 4 Comparison of visual quality of denoised images of denoising networks trained with different data sets. a represents the original star map, b represents the denoised image of the network trained with the data set simulating astronomical noise and stray light, and c represents the denoised image of the network trained with the data set of the present invention.

[0022] Figure 5 Comparison of visual quality of denoised images of denoising networks trained with different data sets. a represents the original star image, b represents the denoised image of the network trained with the simulated data set, and c represents the denoised image of the network trained with the data set of the present invention.

[0023] Figure 6 Comparison of visual quality of denoised images of denoising networks trained with different data sets. a represents the original star image, b represents the denoised image of the network trained with the simulated data set, and c represents the denoised image of the network trained with the data set of the present invention. DETAILED DESCRIPTION

[0024] Specific implementation method 1. Combination Figures 1 to 6 This embodiment describes a method for constructing a positive sample data set for astronomical image denoising applications. This method uses a local maximum detection algorithm with an adaptive threshold to improve the recognition accuracy of star points for the star map to be processed. Subsequently, a geometric relationship model between star points is constructed based on the star point detection results, and efficient star point alignment is achieved through geometric hash matching, thereby ensuring the alignment accuracy and consistency of the star map. Based on the alignment results, areas with different noise and stray light pollution are further cropped to provide high-quality samples for the astronomical image denoising data set, so that the denoising performance of the subsequent model is better and the detail retention effect of the star points is better.

[0025] like Figure 1 As shown, the method for constructing a positive sample data set for astronomical image denoising application described in this embodiment mainly includes the following steps:

[0026] Step 1: Collect and crop star maps at different time periods;

[0027] By observing the same sky area through an astronomical telescope, we can obtain star maps at different times. These star maps contain different degrees of stray light and noise pollution. The same area of ​​the star maps at different times is cropped, and the part with less light pollution is retained first, so as to ensure that the retained image area is relatively clean and has a moderate number of star points.

[0028] In this embodiment, from two star images F1 and F2 of the same sky area taken at different times, the same area is selected for cropping to obtain sub-images G1 and G2. When cropping, the part with less light pollution is retained first, so as to ensure that the retained image area is relatively clean and has a moderate number of star points.

[0029] Step 2: Star point identification and positioning;

[0030] According to each sub-image cropped in step 1, the star points are accurately identified and located through the local maximum detection algorithm, and the coordinate information of the identified star points is retained. Since the star points appear as point-like objects with higher brightness in astronomical images, the potential star point positions can be identified by scanning the neighborhood of the image to find the local maximum value with brightness significantly higher than the surrounding pixels, which has good robustness, especially in more complex backgrounds. For each local maximum point, its pixel coordinates in the image are directly obtained as the precise position of the star point. The coordinates of the star point in sub-image G1 are defined as (x i ,y i ), the star point coordinates in subgraph G2 are defined as (x i ',y i ').

[0031] Step 3: Generation and matching of double triangle structures (generating the topological structure of local star points);

[0032] A double triangle structure is used to represent the geometric relationship between the star points identified in the two sub-images.

[0033] First, four star points are selected and two non-intersecting triangles are constructed between them to form a geometric relationship between the star points.

[0034] Assume that in a picture, there are multiple star points, and the coordinates of the star points are (x i ,y i ), the distances between the four star points are calculated using the following formula:

[0035]

[0036] Among them, d ij represents the Euclidean distance between star point i and star point j.

[0037] Assume that the four star points in subgraph G1 are A(x1,y1), B(x2,y2), C(x3,y3), and D(x4,y4). By comparing the distances between these four star points, the three closest star points A(x1,y1), B(x2,y2), and C(x3,y3) are selected to form the first triangle, and the remaining star point D(x4,y4) automatically becomes a vertex in the second triangle.

[0038] Then, by calculating the distance from the remaining star point D(x4,y4) to each vertex of the first triangle, two vertices with shorter distances are selected to ensure that the second triangle is not too distorted and the geometric relationship between the star points is more stable and natural. Calculate the distance from point D(x4,y4) to each vertex in triangle ABC:

[0039]

[0040] Select two points with a shorter distance, assuming they are A and B, then the second triangle is DAB. Similarly, we can assume that the four star points in subgraph G2 are E, F, G, and H. The three closest points E, F, and G form the first triangle EFG, and the remaining star point H is the vertex and the two closest points E and F form the second triangle HEF.

[0041] In both subgraphs, two triangles are formed by four star points, and their side lengths, angles and other geometric features are recorded respectively. By calculating the lengths of the three sides, the shape features of the triangle can be obtained.

[0042]

[0043] Using the law of cosines, we can calculate the size of each angle. For example, the angle θ A The calculation formula is:

[0044]

[0045] Similarly, the angle θ B and θ C It can also be calculated using the law of cosines.

[0046] The area of ​​a triangle can also be used as a geometric feature. By calculating the area, the shape of the triangle can be further described. The area is calculated using Heron's formula, as follows:

[0047]

[0048] Similarly, the area Area2 of the triangle DAB can be obtained.

[0049] These two triangles constitute the local topological relationship between the star points. After extracting the geometric features of the two triangles, a unique geometric hash code is generated through the following steps for matching the star points.

[0050] All geometric features (side lengths, angles, areas) are combined into a feature vector that fully describes the geometry of each triangle:

[0051] F=[L AB , L BC , L CA ,θ A ,θ B ,θ C ,Area1,L DA , L AB , L BD ,θ D ,θ A ,θ B , Area2]

[0052] In order to generate hash codes, it is necessary to discretize the continuous features, divide each eigenvalue into several discrete intervals, and quantize the values ​​in the eigenvector. Assuming that each eigenvalue is quantized into k intervals, then each eigenvalue f i is mapped to a discrete eigenvalue d i ,in:

[0053] d i =quantize(f i , k)

[0054] The discretized eigenvalues ​​are combined into a unique hash code. Assume there are n discretized eigenvalues ​​d1, d2, ..., d n , then the final geometric hash code can be expressed as:

[0055]

[0056] In this way, the hash code maps the different geometric features of the star points into a unique integer value. By matching similar hash values, the corresponding star point sets in the two images can be quickly found. Assuming that the hash codes in the two images are H1 and H2, respectively, the matching process can be judged by the following formula:

[0057] |H1-H2|<∈

[0058] Among them, ∈ is the set tolerance range, which is used to allow some slight error matching. If the conditions are met, the star point sets represented by the two hash codes are considered to match.

[0059] Step 4: Align the star map;

[0060] After matching the star points obtained in step 3, it is necessary to perform an affine transformation on the star point coordinates of the two images so that they can be correctly aligned on the original image and ensure that the corresponding star points in the two sub-images can accurately overlap in the original coordinate system.

[0061] Assume that a set of matching star points has been found in two images. Set the coordinates of the star points in the first image to (x1, y1), (x2, y2), ..., (x n ,y n ), the coordinates of the matching star points corresponding to the second image are (x′1, y′1), (x′2, y′2), …, (x′ n , y′ n ). After obtaining the matched star points, the star point coordinates of the two images need to be transformed so that they can be correctly aligned on the original image. The calculation of the affine matrix is ​​usually based on a set of matched star point pairs. The optimal transformation matrix is ​​found by solving the least squares problem to keep the geometric relationship between the two sets of star points consistent.

[0062] Affine transformation can handle linear transformations such as scaling, rotation, and translation to ensure that the corresponding star points in the two sub-images can accurately overlap in the original coordinate system. Affine transformation can be expressed in the following matrix form:

[0063]

[0064] Where (x i ,y i ) is the coordinate of the star point in the first picture, (x′ i, y′ i ) are the coordinates of the corresponding star point in the second image, a, b, c, d are the parameters of the affine transformation matrix, t x and t y is the translation amount.

[0065] In order to calculate the six parameters a, b, c, d, tx, ty in the affine matrix, at least three pairs of matching star points are required. For each pair of matching star points, the following linear equation can be written:

[0066] x′ i =ax i +by i +t x

[0067] y′ i =cx i +dy i +t y

[0068] Substituting each pair of matching star points into the equation, a linear system of equations can be obtained. For n pairs of matching star points (n ≥ 3), the system of equations can be written as:

[0069]

[0070] X′=AP

[0071] Among them, X' is the coordinate of the matching target star point, A is the known star point coordinate matrix, and P is the affine matrix parameter to be solved.

[0072] The least squares method is used to solve the equations and find the best affine matrix parameters a, b, c, d, tx, ty to minimize the matching error. The minimized objective function is:

[0073]

[0074] By solving this minimization problem, we can obtain the affine transformation matrix P, which transforms all the star points in the second image and maps them to the coordinate system of the first image. For any star point (x, y), the coordinates after transformation by the affine matrix are:

[0075]

[0076] Through this transformation, the matching areas in the star images taken at two different times can be well aligned.

[0077] Step 5: Selection of positive and negative samples of the data set;

[0078] After the star map is aligned in step 4, the unpolluted (or weakly polluted) area and the strongly polluted area in the star map can be further cropped to generate a data set. The strongly polluted area contains parts with excessive brightness or excessive background noise as negative samples in the data set, while the unpolluted or weakly polluted area preserves relatively realistic star points and backgrounds as positive samples in the data set for subsequent denoising model training. Through this star map data taken at different time periods in real scenes, an astronomical image denoising data set covering various pollution types can be constructed to improve the generalization performance and denoising effect of the model under actual observation conditions.

[0079] Specific implementation method 2: Combination Figures 1 to 6 This embodiment is described as an example of a method for constructing a positive sample data set for astronomical image denoising application described in the first embodiment.

[0080] 1. Working conditions;

[0081] This experiment uses an Intel core i7-7700K CPU @ 4.20 GHz * 8 processor, a PC running Windows 10, a graphics card is a GeForce GTX 1070Ti, and the programming language is Python.

[0082] 2. Experimental content and result analysis;

[0083] like Figure 2 The figure shows the process of generating the topological structure of local star points and realizing star map alignment, which briefly summarizes the subdivision process. Figure 3 As shown, when the present invention realizes star map matching, the star points in the image are identified by a local maximum detection algorithm to better generate a suitable double triangle structure; after matching the double triangle structure, the star map is matched and aligned using an affine transformation matrix, so that the star map matching efficiency is greatly improved.

[0084] In order to better compare the performance of the dataset construction method proposed in this paper on star image denoising and stray light removal, a comparative analysis will be conducted from the perspective of subjective visual quality.

[0085] like Figure 4 As shown, by comparing the original star map, it can be seen that the denoising effect of the dataset construction method of the present invention after training the network has better subjective visual quality than the denoising effect of the dataset constructed by simulating astronomical noise and stray light after training the network, and there will be no phenomenon of missing star points, which can be close to the original image to a great extent. Figure 5 , Figure 6 As shown, the denoising network is trained on the data set of the present invention and the simulated data set respectively to denoise the star map. By comparing with the original star map, it can be found that the method of the present invention can better preserve and reconstruct the details and brightness information of the targets in the star map compared with the method of the simulated data set.

[0086] pass Figures 4 to 6 It can be seen that when performing star image denoising, the present invention can better preserve the detail information and brightness information of space targets, and is superior to other methods in terms of subjective visual quality.

[0087] The above experimental results show that for the star map to be processed, the present invention improves the star point recognition accuracy through the local maximum star point detection algorithm, and subsequently adopts a double triangle structure to establish the geometric relationship between the star points. By selecting star points to construct two non-intersecting triangles, the side length and angle features are extracted to generate a geometric hash code, thereby achieving efficient and accurate star point matching. The use of a double triangle structure can make full use of the geometric characteristics between star points and enhance the robustness and noise resistance of matching. Based on the matching results, the star map can be further accurately aligned and cropped, and an astronomical image denoising data set covering a variety of pollution types can be constructed. Ultimately, the processed star map has a better denoising effect and better retention of star point details.

[0088] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for constructing a positive sample dataset for astronomical image denoising applications, characterized by: The method is implemented by the following steps: Step 1: obtaining star maps of different time periods, selecting the same area of ​​the star maps for cropping, and obtaining two sub-images; Step 2: Star point identification and positioning; For each cropped sub-image, star point identification and positioning are performed through local maximum detection; Step 3: Generation and matching of double triangle structures; The star points identified in the two sub-images are represented by a double triangle structure to represent the geometric relationship between the star points; the corresponding star points in the two sub-images are matched; Step 4: According to the result of star point matching in step 3, complete the star map alignment; Step 5: Selection of positive and negative samples of the data set; According to the results of star map alignment in step 4, the unpolluted areas and strongly polluted areas in the star map are cropped to generate a data set.

2. The method for constructing a positive sample dataset for astronomical image denoising application according to claim 1, characterized in that: In step 1, two star maps F1 and F2 are obtained from the same sky area photographed at different time periods, and the same regions of the two star maps F1 and F2 are selected for cropping to obtain sub-maps G1 and G2.

3. The method for constructing a positive sample dataset for astronomical image denoising application according to claim 2, characterized in that: In each cropped sub-image, the star points are accurately identified and located through the local maximum detection algorithm, and the coordinate information of the identified star points is retained.

4. The method for constructing a positive sample dataset for astronomical image denoising application according to claim 3, characterized in that: Scan the neighborhood of the sub-image to find the local maximum value with brightness higher than the surrounding pixels, identify the potential star point position, and directly obtain the pixel coordinates in the image as the precise position of the star point for each local maximum point. The star point coordinates in sub-image G1 are (x i ,y i ), the star point coordinates in subgraph G2 are (x i ',y i ').

5. The method for constructing a positive sample dataset for astronomical image denoising application according to claim 1, characterized in that: The specific process of step three is: Four star points are selected and two non-intersecting triangles are constructed between them to form a geometric relationship between the star points. In each triangle, the side length and angle features are extracted to generate a geometric hash code for matching the star points. After the hash code is generated, the corresponding star point sets in the two images are found by matching similar hash values.

6. The method for constructing a positive sample dataset for astronomical image denoising application according to claim 5, characterized in that: First, the four star points in subgraph G1 are set to be A, B, C, and D. By comparing the distances between the four star points, the three closest star points A, B, and C are selected to form the first triangle. Star point D automatically becomes a vertex in the second triangle. The distance from star point D to each vertex of the first triangle is determined by selecting two vertices with shorter distances. Set them to be A and B. Then the second triangle is DAB. Similarly, let the four star points in subgraph G2 be E, F, G, and H. The three closest points E, F, and G form the first triangle EFG. The remaining star point H is the vertex and the two closest points E and F form the second triangle HEF. Then, in both sub-images, double triangles are formed through four star points, and their geometric features of side length, angle and area are recorded respectively; After extracting the geometric features of the double triangle, a unique geometric hash code is generated for star point matching; Finally, by matching similar hash values, we find the corresponding star point sets in the two sub-images. We set the hash codes in the two sub-images to be H1 and H2 respectively, and use the following formula to determine: ∣H1-H2∣<∈ Among them, ∈ is the set tolerance range. If the condition is met, the star point sets represented by the two hash codes are considered to match.

7. The method for constructing a positive sample dataset for astronomical image denoising application according to claim 1, characterized in that: The specific process of step 4 is as follows: Perform affine transformation on the star point coordinates of the two sub-images, calculate the affine matrix parameters using the least squares method, and apply affine transformation to align the star images.

8. The method for constructing a positive sample data set for astronomical image denoising application according to claim 7, characterized in that: Set a set of matching star points in two sub-images; set the star point coordinates of the first sub-image to (x1, y1), (x2, y2),…, (x n ,y n ), the coordinates of the matching star points corresponding to the second sub-image are (x′1, y′1), (x′2, y′2),…, (x′ n ,y′ n ); After obtaining the matching star points, the star point coordinates of the two sub-images are transformed. The affine transformation uses the following matrix form: Where (x i ,y i ) is the coordinate of the star point in the first sub-image, (x′ i ,y′ i ) are the coordinates of the corresponding star point in the second sub-image, a, b, c, d are the parameters of the affine transformation matrix, t x and t y is the translation amount; Substituting each pair of matching star points into the following equation, we obtain a system of linear equations; x′ i =ax i +by i +t x the' i =cx i +to i +t y For n pairs of star point matching, n ≥ 3, the system of equations is written as: X′=AP Where X' is the coordinate of the target star point to be matched, A is the known star point coordinate matrix, and P is the affine matrix parameter to be solved; The least squares method is used to solve the equations and find the optimal affine matrix parameters a, b, c, d, tx, ty to minimize the matching error. The minimized objective function is: By solving the minimization objective function, the affine transformation matrix P is obtained, and all the star points in the second sub-image are transformed and mapped to the coordinate system of the first sub-image; for any star point (x, y), the coordinates after affine matrix transformation are (x′, y′), and the matching areas in the star images taken at two different times are aligned after affine transformation.

Images