CN117974571A - Remote sensing image evaluation model training method, remote sensing image evaluation method and device - Google Patents
Remote sensing image evaluation model training method, remote sensing image evaluation method and device Download PDFInfo
- Publication number
- CN117974571A CN117974571A CN202311862159.XA CN202311862159A CN117974571A CN 117974571 A CN117974571 A CN 117974571A CN 202311862159 A CN202311862159 A CN 202311862159A CN 117974571 A CN117974571 A CN 117974571A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing image
- sample
- quality
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000012549 training Methods 0.000 title claims abstract description 45
- 238000013210 evaluation model Methods 0.000 title claims abstract description 41
- 238000011156 evaluation Methods 0.000 title claims abstract description 27
- 238000012545 processing Methods 0.000 claims abstract description 29
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 18
- 238000013441 quality evaluation Methods 0.000 claims abstract description 13
- 238000010606 normalization Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 3
- 238000013077 scoring method Methods 0.000 abstract description 4
- 238000013135 deep learning Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The application relates to a remote sensing image evaluation model training method, a remote sensing image evaluation method and a remote sensing image evaluation device, wherein the remote sensing image evaluation model training method comprises the following steps: acquiring a sample high-quality remote sensing image, and performing digital image processing on the sample high-quality remote sensing image to generate a sample low-quality remote sensing image; generating a sample difference map based on the sample high-quality remote sensing image and the sample low-quality remote sensing image; inputting the sample difference graph and the corresponding sample label into a pre-constructed convolutional neural network for training to obtain a remote sensing image evaluation model; the sample label is used for representing image quality evaluation of the sample low-quality remote sensing image. And obtaining a remote sensing image evaluation model for evaluating the image quality of the remote sensing image. Compared with the traditional artificial subjective scoring method, the method realizes automatic evaluation of remote sensing image quality through deep learning of the convolutional neural network, and effectively improves working efficiency.
Description
Technical Field
The application relates to the technical field of remote sensing images, in particular to a remote sensing image evaluation model training method, a remote sensing image evaluation method and a remote sensing image evaluation device.
Background
The remote sensing image is an important space information acquisition source and plays an important role in the aspects of national economic construction, national defense safety and the like. However, in various industrial applications, the quality of the remote sensing image directly affects the data utilization rate and the processing efficiency, and therefore, the quality of the acquired remote sensing image needs to be evaluated before use.
At present, an artificial subjective scoring method is often adopted to evaluate the quality of the remote sensing image. The score is given by manual subjective judgment by formulating quantitative scoring criteria such as brightness, contrast, color deviation, object discernability, etc.
However, this method is labor-intensive and difficult to use in a large number of remote sensing data services.
Disclosure of Invention
In view of the above, the application provides a remote sensing image evaluation model training method, a remote sensing image evaluation method and a remote sensing image evaluation device.
According to an aspect of the present application, there is provided a remote sensing image evaluation model training method, including:
acquiring a sample high-quality remote sensing image, and performing digital image processing on the sample high-quality remote sensing image to generate a sample low-quality remote sensing image;
generating a sample difference map based on the sample high-quality remote sensing image and the sample low-quality remote sensing image;
Inputting the sample difference graph and the corresponding sample label into a pre-constructed convolutional neural network for training to obtain a remote sensing image evaluation model; the sample label is used for representing image quality evaluation of the sample low-quality remote sensing image.
In one possible implementation manner, when the sample high-quality remote sensing image is processed digitally to generate the sample low-quality remote sensing image, at least one method of changing the image contrast of the sample high-quality remote sensing image, changing the image brightness of the sample high-quality remote sensing image and adding gaussian noise to the sample high-quality remote sensing image is adopted.
In one possible implementation, the image contrast of the sample high quality remote sensing image is changed by randomly generating contrast parameters;
changing the image contrast of the sample high-quality remote sensing image by randomly generating brightness parameters;
And adding Gaussian noise to the sample high-quality remote sensing image through randomly generating Gaussian noise parameters.
In one possible implementation, the sample tag is constructed based on at least one of a sharpness bias, a brightness bias, a contrast bias, and a color bias between the sample low quality remote sensing image and the sample high quality remote sensing image.
In one possible implementation manner, when the sample difference map is generated based on the sample high-quality remote sensing image and the sample low-quality remote sensing image, the sample difference map is obtained by performing image normalization processing on the sample high-quality remote sensing image and the sample low-quality remote sensing image.
In one possible implementation manner, when performing image normalization processing on the sample high-quality remote sensing image and the sample low-quality remote sensing image, at least one of hue deviation, saturation deviation and brightness deviation between the sample high-quality remote sensing image and the sample low-quality remote sensing image is performed.
According to another aspect of the present application, there is provided a remote sensing image evaluation method, including:
Acquiring a remote sensing image to be evaluated and a corresponding high-quality reference image;
generating a difference image based on the remote sensing image to be evaluated and the high-quality reference image;
And inputting the difference map into a remote sensing image evaluation model trained by any one of the methods to obtain the quality score of the remote sensing image to be evaluated.
In one possible implementation, the corresponding high-quality reference image is acquired based on the geographic location characterized by the remote sensing image to be evaluated.
According to another aspect of the present application, there is provided a remote sensing image evaluation model training apparatus, comprising: the device comprises a sample processing module, a sample generating module and a training module;
The sample processing module is configured to acquire a sample high-quality remote sensing image, and perform digital image processing on the sample high-quality remote sensing image to generate a sample low-quality remote sensing image;
The sample generation module is configured to generate a sample difference map based on the sample high-quality remote sensing image and the sample low-quality remote sensing image;
the training module is configured to input the sample difference graph and the corresponding sample label into a pre-constructed convolutional neural network for training to obtain a remote sensing image evaluation model; the sample label is used for representing image quality evaluation of the sample low-quality remote sensing image.
According to another aspect of the present application, there is provided a remote sensing image evaluation apparatus including: the device comprises a reference image acquisition module, a difference image generation module and an evaluation module;
The reference image acquisition module is configured to acquire a remote sensing image to be evaluated and a corresponding high-quality reference image;
The difference map generation module is configured to generate a difference map based on the remote sensing image to be evaluated and the high-quality reference image;
The evaluation module is configured to input the difference map into a remote sensing image evaluation model obtained through training by any one of the methods, so as to obtain the quality score of the remote sensing image to be evaluated.
The method is suitable for obtaining the remote sensing image evaluation model based on convolutional neural network training, and the obtained remote sensing image evaluation model is used for evaluating the image quality of the remote sensing image. And carrying out digital image processing on the sample high-quality remote sensing image with higher image to generate a corresponding sample low-quality remote sensing image with lower image quality, wherein the generated sample low-quality remote sensing image is used for simulating the actually acquired remote sensing image. The sample difference map and the sample label constructed based on the sample high-quality remote sensing image and the sample low-quality image are respectively used for representing the difference degree of image quality between the sample high-quality remote sensing image and the sample low-quality image and evaluating the image quality of the sample low-quality remote sensing image, namely, the larger the difference degree is, the lower the quality evaluation is. And inputting the sample difference graph and the corresponding sample label into a pre-constructed convolutional neural network for training to obtain a remote sensing image evaluation model for evaluating the image quality of the remote sensing image. Compared with the traditional artificial subjective scoring method, the method realizes automatic evaluation of remote sensing image quality through deep learning of the convolutional neural network, and effectively improves working efficiency.
Other features and aspects of the present application will become apparent from the following detailed description of exemplary embodiments, which proceeds with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate exemplary embodiments, features and aspects of the application and together with the description, serve to explain the principles of the application.
FIG. 1 shows a flowchart of a remote sensing image evaluation model training method according to an embodiment of the present application;
FIG. 2 shows a flow chart of a remote sensing image evaluation method according to an embodiment of the present application;
FIG. 3 shows an overall flowchart of a remote sensing image evaluation model training method and a remote sensing image evaluation method of an embodiment of the present application;
Fig. 4 is a main body structure diagram of a remote sensing image evaluation model training device according to an embodiment of the present application;
Fig. 5 is a main block diagram of a remote sensing image evaluation apparatus according to an embodiment of the present application.
Detailed Description
Various exemplary embodiments, features and aspects of the application will be described in detail below with reference to the drawings. In the drawings, like reference numbers indicate identical or functionally similar elements. Although various aspects of the embodiments are illustrated in the accompanying drawings, the drawings are not necessarily drawn to scale unless specifically indicated.
The word "exemplary" is used herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.
In addition, numerous specific details are set forth in the following description in order to provide a better illustration of the application. It will be understood by those skilled in the art that the present application may be practiced without some of these specific details. In some instances, well known methods, procedures, components, and circuits have not been described in detail so as not to obscure the present application.
For the convenience of understanding the technical scheme of the present application, the terms in the present application will be explained correspondingly.
Fig. 1 shows a flowchart of a remote sensing image evaluation model training method according to an embodiment of the present application. As shown in fig. 1, the remote sensing image evaluation model training method includes: step S100: obtaining a sample high-quality remote sensing image, wherein the sample high-quality remote sensing image is a remote sensing image obtained under the conditions of no cloud and no fog, no speckle interference and normal sensor state, and performing digital image processing on the sample high-quality remote sensing image to generate a sample low-quality remote sensing image; step S200: generating a sample difference map based on the sample high-quality remote sensing image and the sample low-quality remote sensing image; step S300: inputting the sample difference graph and the corresponding sample label into a pre-constructed convolutional neural network for training to obtain a remote sensing image evaluation model; the sample label is used for representing image quality evaluation of the sample low-quality remote sensing image.
The method is suitable for obtaining the remote sensing image evaluation model based on convolutional neural network training, and the obtained remote sensing image evaluation model is used for evaluating the image quality of the remote sensing image. And carrying out digital image processing on the sample high-quality remote sensing image with higher image to generate a corresponding sample low-quality remote sensing image with lower image quality, wherein the generated sample low-quality remote sensing image is used for simulating the actually acquired remote sensing image. The sample difference map and the sample label constructed based on the sample high-quality remote sensing image and the sample low-quality image are respectively used for representing the difference degree of image quality between the sample high-quality remote sensing image and the sample low-quality image and evaluating the image quality of the sample low-quality remote sensing image, namely, the larger the difference degree is, the lower the quality evaluation is. And inputting the sample difference graph and the corresponding sample label into a pre-constructed convolutional neural network for training to obtain a remote sensing image evaluation model for evaluating the image quality of the remote sensing image. Compared with the traditional artificial subjective scoring method, the method realizes automatic evaluation of remote sensing image quality through deep learning of the convolutional neural network, and effectively improves working efficiency.
The method is characterized in that the sample difference map data are input into a convolutional neural network and training is started, weight parameters in the network are updated through a back propagation algorithm in the training process, and iteration is continued until the accuracy is more than 85%, so that a remote sensing image evaluation model is obtained.
In one possible implementation manner, a high-quality remote sensing image with higher image quality is selected as a sample from the existing remote sensing images by means of manual selection. When digital image processing is carried out on the sample high-quality remote sensing image to generate a sample low-quality remote sensing image, at least one method of changing the image contrast of the sample high-quality remote sensing image, changing the image brightness of the sample high-quality remote sensing image and adding Gaussian noise to the sample high-quality remote sensing image is adopted.
The image contrast of the high-quality remote sensing image of the sample is changed by randomly generating contrast parameters; changing the image contrast of the sample high-quality remote sensing image by randomly generating brightness parameters; and adding Gaussian noise to the sample high-quality remote sensing image through randomly generating Gaussian noise parameters.
It should be noted that, when the contrast parameter, the brightness parameter and the gaussian noise parameter are randomly generated, a person skilled in the art can select a range of the randomly generated parameters according to actual situations, so that the generated sample low-quality remote sensing image is more similar to the remote sensing image actually acquired under different situations.
Preferably, when digital image processing is performed on a sample high-quality remote sensing image, the value range of a randomly generated contrast parameter (pixel value multiplied by the parameter) is 0.2 to 0.5, the value range of a randomly generated brightness parameter (pixel value added with the parameter) is 0 to 128, and the value of a gaussian distribution with a mean value of 1 and a variance of 1 is randomly generated.
Further, the high-quality remote sensing image of the sample is modified by utilizing the contrast parameter, the brightness parameter and the Gaussian noise parameter which are randomly generated, and the corresponding low-quality remote sensing image of the sample is obtained. It should be noted that, when the sample high-quality remote sensing image is modified, only the contrast parameter and the brightness parameter are modified, and the gaussian noise parameter is added, so that the image size is not modified. The high-quality remote sensing images of the samples and the low-quality remote sensing images of the corresponding generated samples form a pair of samples, and the samples are used for training a convolutional neural network which is built in advance.
After obtaining the sample high-quality remote sensing image and the corresponding sample low-quality remote sensing image by using any one of the methods, constructing a sample label for representing the image quality evaluation of the sample low-quality remote sensing image, wherein the sample label is constructed based on the sample low-quality remote sensing image and the corresponding sample high-quality remote sensing image, that is, the image quality evaluation of the sample low-quality remote sensing image is based on the sample high-quality remote sensing image.
The sample label adopts a scoring system, namely, the higher the image quality of the sample low-quality remote sensing image is, the higher the scoring score is, the lower the image quality is, and the lower the scoring score is.
Preferably, the score of the sample tag is 0 to 5.
Further, the sample label is constructed based on at least one of a definition deviation, a brightness deviation, a contrast deviation and a color deviation between the sample low-quality remote sensing image and the sample high-quality remote sensing image, that is, the smaller the definition deviation between the two is, the higher the scoring score of the sample label is; the smaller the brightness deviation between the two, the higher the scoring score of the sample label; the smaller the contrast deviation between the two, the higher the scoring score of the sample label; the smaller the color deviation between the two, the higher the scoring score of the sample label.
Further, based on the definition deviation between the sample low-quality remote sensing image and the sample high-quality remote sensing image, a definition score is obtained, and the smaller the definition deviation is, the higher the definition score is; obtaining a brightness score based on brightness deviation between the sample low-quality remote sensing image and the sample high-quality remote sensing image, wherein the brightness score is higher when the brightness deviation is smaller; obtaining a contrast score based on contrast deviation between the sample low-quality remote sensing image and the sample high-quality remote sensing image, wherein the smaller the contrast deviation is, the higher the contrast score is; and obtaining a color score based on the color deviation between the sample low-quality remote sensing image and the sample high-quality remote sensing image, wherein the smaller the color deviation is, the higher the color score is.
And calculating the score of the sample label based on the definition score, the brightness score, the contrast score and the color score obtained by any one of the methods.
Preferably, the score of the sharpness score, the brightness score, the contrast score and the color score are all 0 to 5, and the average value of the sharpness score, the brightness score, the contrast score and the color score is taken as the score of the final sample label.
It should be noted that, when the sample label is constructed based on the sharpness deviation, brightness deviation, contrast deviation and color deviation between the sample low-quality remote sensing image and the sample high-quality remote sensing image, the scoring is performed manually, i.e. the scoring scores of the sharpness score, the brightness score, the contrast score and the color score are given by the manual, respectively, and then the average value is taken as the scoring score of the final sample label.
TABLE 1
And inputting the sample label obtained by the method and the corresponding sample difference map into a pre-constructed convolutional neural network to train the sample label to obtain a remote sensing image evaluation model, wherein the sample difference map and the corresponding sample label are constructed based on the same pair of sample high-quality remote sensing images and sample low-quality remote sensing images.
When a sample difference image is generated based on a sample high-quality remote sensing image and a sample low-quality remote sensing image, the sample difference image is obtained by carrying out image normalization processing on the sample high-quality remote sensing image and the sample low-quality remote sensing image. And constructing a sample high-quality remote sensing image and a sample low-quality remote sensing image into a sample difference map through image normalization processing.
Further, when the sample high-quality remote sensing image and the sample low-quality remote sensing image are subjected to image normalization processing, the image normalization processing is performed based on at least one of hue deviation, saturation deviation and brightness deviation between the sample high-quality remote sensing image and the sample low-quality remote sensing image.
That is, when performing the image normalization processing on the sample high-quality remote sensing image and the sample low-quality remote sensing image, the sample high-quality remote sensing image and the sample low-quality remote sensing image are first respectively transformed into an HIS (Hue-Saturation-Intensity) color space, so as to obtain Hue values, saturation values and Intensity values of the sample high-quality remote sensing image and the sample low-quality remote sensing image, and a sample difference map is constructed based on the difference value of the Hue values, the difference value of the Saturation values and the difference value of the Intensity values of the two.
In one possible implementation manner, the sample difference map obtained by performing image normalization processing on the sample high-quality remote sensing image and the sample low-quality remote sensing image is an RGB three-channel image.
Further, the difference value of hue values, the difference value of saturation values and the difference value of intensity values between the sample high-quality remote sensing image and the sample low-quality remote sensing image are respectively used as RGB three channels of a sample difference image, and then the sample difference image is constructed and obtained.
Specifically, calculating the difference of H (hue) values of each pixel point of a sample high-quality remote sensing image and a sample low-quality remote sensing image, and carrying out normalization processing on the obtained difference of the H values of each pixel point to normalize the difference to 0-255, wherein the difference of each pixel point after normalization is used as the value of an R channel corresponding to each pixel point in a sample difference graph; similarly, the difference of the S (saturation) values of each pixel point of the sample high-quality remote sensing image and the sample low-quality remote sensing image is normalized to be between 0 and 255 and is used as the value of the G channel corresponding to each pixel point in the sample difference graph; and normalizing the difference of the I (brightness) values of each pixel point of the sample high-quality remote sensing image and the sample low-quality remote sensing image to be between 0 and 255, and taking the difference as the value of a B channel corresponding to each pixel point in the sample difference graph. Therefore, values of an R channel, a G channel and a B channel of each pixel point in the sample difference graph are obtained based on the method, and a final sample difference graph is obtained.
And inputting the sample difference graph and the corresponding sample label obtained by any one of the methods into a convolutional neural network constructed in advance for training to obtain a final remote sensing image evaluation model.
According to another aspect of the present application, there is provided a remote sensing image evaluation method, as shown in fig. 2, the method comprising: step S001: acquiring a remote sensing image to be evaluated and a corresponding high-quality reference image, wherein a plurality of real remote sensing image groups with similar time phases (the common time deviation is less than 7 days) and the same resolution are acquired, and selecting a plurality of images of the same area by adopting a manual screening method, wherein the remote sensing image to be evaluated is the image with the imaging condition crossed, and the high-quality reference image is the image with the high-quality condition meeting the training sample; step S002: generating a difference image based on the remote sensing image to be evaluated and the high-quality reference image; step S003: inputting the difference image into a remote sensing image evaluation model obtained by training by any one of the methods, performing operation of a convolution layer and a pooling layer, extracting edge, texture and color features from an original image by the convolution layer through convolution operation and an activation function, flattening the image into a feature vector after processing of a plurality of convolution layers and pooling layers, performing regression on the feature vector by a full-connection layer, and finally outputting the quality score of the remote sensing image to be evaluated.
The method comprises the steps of obtaining a corresponding high-quality reference image based on a remote sensing image to be evaluated, generating a difference image based on the remote sensing image to be evaluated and the high-quality reference image, wherein the step of generating the difference image is similar to the step of generating a sample difference image, namely, the difference value based on the hue value difference value, the saturation value difference value and the intensity value difference value between the remote sensing image to be evaluated and the high-quality reference image are respectively used as RGB three channels of the difference image, so that the difference image is constructed and obtained, and the detailed description is omitted. And inputting the generated difference graph into a remote sensing image evaluation model obtained by the training, and further obtaining the quality score of the remote sensing image to be evaluated, namely the score of the sample label. Compared with the existing digital image processing method, namely, parameters such as peak signal-to-noise ratio, structural similarity, image entropy, gradient and the like of the remote sensing image to be evaluated are directly calculated, quality scores are obtained as image quality evaluation indexes, and the parameters are extracted only from the angle of the information of the image itself, but the characteristics of different images are different, so that various indexes cannot be completely related to the image quality, and a certain deviation exists in the evaluation result. According to the method, the high-quality reference image is introduced to serve as a reference for quality evaluation, so that the quality scoring accuracy of the remote sensing image to be evaluated is improved.
When the corresponding high-quality reference image of the remote sensing image to be evaluated is obtained, the method is carried out based on the geographic position represented by the remote sensing image to be evaluated.
Further, based on the geographic positions represented by the remote sensing images to be evaluated, searching the high-quality remote sensing images with the same represented geographic positions in a preset reference image library to serve as high-quality reference images.
Furthermore, after the high-quality remote sensing image which is the same as the geographical position represented by the remote sensing image to be evaluated is obtained in the preset reference gallery, preprocessing is further included on the obtained high-quality remote sensing image, and then the corresponding high-quality reference image is obtained.
The method for preprocessing the high-quality remote sensing image comprises the following steps: resampling the high-quality remote sensing image based on the resolution of the remote sensing image to be evaluated, so that the size of the high-quality remote sensing image is identical to the size of the remote sensing image to be evaluated.
And the difference image is input into a remote sensing image evaluation model obtained through training by any one of the methods, so that the quality score of the remote sensing image to be evaluated is obtained.
Still further, referring to fig. 4, according to another aspect of the present application, there is also provided a remote sensing image evaluation model training apparatus 100, including: a sample processing module 110, a sample generation module 120, and a training module 130. The sample processing module 110 is configured to acquire a sample high-quality remote sensing image, and perform digital image processing on the sample high-quality remote sensing image to generate a sample low-quality remote sensing image; a sample generation module 120 configured to generate a sample disparity map based on the sample high quality remote sensing image and the sample low quality remote sensing image; the training module 130 is configured to input the sample difference graph and the corresponding sample label into a pre-constructed convolutional neural network for training to obtain a remote sensing image evaluation model; the sample label is used for representing image quality evaluation of the sample low-quality remote sensing image.
Referring to fig. 5, according to another aspect of the present application, there is also provided a remote sensing image evaluation apparatus 200, including: a reference image acquisition module 210, a disparity map generation module 220, and an evaluation module 230; a reference image acquisition module 210 configured to acquire a remote sensing image to be evaluated and a corresponding high-quality reference image; a difference map generation module 220 configured to generate a difference map based on the remote sensing image to be evaluated and the high quality reference image; the evaluation module 230 is configured to input the difference map into the remote sensing image evaluation model obtained by training by any one of the methods, so as to obtain a quality score of the remote sensing image to be evaluated.
The foregoing description of embodiments of the application has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the various embodiments described. The terminology used herein was chosen in order to best explain the principles of the embodiments, the practical application, or the technical improvements in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.
Claims (10)
1. The remote sensing image evaluation model training method is characterized by comprising the following steps of:
acquiring a sample high-quality remote sensing image, and performing digital image processing on the sample high-quality remote sensing image to generate a sample low-quality remote sensing image;
generating a sample difference map based on the sample high-quality remote sensing image and the sample low-quality remote sensing image;
Inputting the sample difference graph and the corresponding sample label into a pre-constructed convolutional neural network for training to obtain a remote sensing image evaluation model; the sample label is used for representing image quality evaluation of the sample low-quality remote sensing image.
2. The method of claim 1, wherein at least one of changing an image contrast of the sample high quality remote sensing image, changing an image brightness of the sample high quality remote sensing image, and adding gaussian noise to the sample high quality remote sensing image is employed in digitally image processing the sample high quality remote sensing image to generate the sample low quality remote sensing image.
3. The method of claim 2, wherein the image contrast of the sample high quality remote sensing image is changed by randomly generating contrast parameters;
changing the image contrast of the sample high-quality remote sensing image by randomly generating brightness parameters;
And adding Gaussian noise to the sample high-quality remote sensing image through randomly generating Gaussian noise parameters.
4. The method of claim 1, wherein the sample tag is constructed based on at least one of a sharpness bias, a brightness bias, a contrast bias, and a color bias between the sample low quality remote sensing image and the sample high quality remote sensing image.
5. The method according to any one of claims 1 to 4, wherein the sample difference map is generated based on the sample high-quality remote sensing image and the sample low-quality remote sensing image by performing an image normalization process on the sample high-quality remote sensing image and the sample low-quality remote sensing image.
6. The method of claim 5, wherein the performing the image normalization process on the sample high quality remote sensing image and the sample low quality remote sensing image is performed based on at least one of a hue deviation, a saturation deviation, and a brightness deviation between the sample high quality remote sensing image and the sample low quality remote sensing image.
7. The remote sensing image evaluation method is characterized by comprising the following steps of:
Acquiring a remote sensing image to be evaluated and a corresponding high-quality reference image;
generating a difference image based on the remote sensing image to be evaluated and the high-quality reference image;
Inputting the difference map into a remote sensing image evaluation model obtained through training by the method of any one of claims 1 to 6 to obtain the quality score of the remote sensing image to be evaluated.
8. The method of claim 7, wherein the corresponding high quality reference image is obtained based on a geographic location characterized by the remote sensing image to be evaluated.
9. A remote sensing image evaluation model training device, characterized by comprising: the device comprises a sample processing module, a sample generating module and a training module;
The sample processing module is configured to acquire a sample high-quality remote sensing image, and perform digital image processing on the sample high-quality remote sensing image to generate a sample low-quality remote sensing image;
The sample generation module is configured to generate a sample difference map based on the sample high-quality remote sensing image and the sample low-quality remote sensing image;
the training module is configured to input the sample difference graph and the corresponding sample label into a pre-constructed convolutional neural network for training to obtain a remote sensing image evaluation model; the sample label is used for representing image quality evaluation of the sample low-quality remote sensing image.
10. A remote sensing image evaluation device, comprising: the device comprises a reference image acquisition module, a difference image generation module and an evaluation module;
The reference image acquisition module is configured to acquire a remote sensing image to be evaluated and a corresponding high-quality reference image;
The difference map generation module is configured to generate a difference map based on the remote sensing image to be evaluated and the high-quality reference image;
the evaluation module is configured to input the difference map into a remote sensing image evaluation model obtained through training by the method of any one of claims 1 to 6, so as to obtain a quality score of the remote sensing image to be evaluated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311862159.XA CN117974571A (en) | 2023-12-29 | 2023-12-29 | Remote sensing image evaluation model training method, remote sensing image evaluation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311862159.XA CN117974571A (en) | 2023-12-29 | 2023-12-29 | Remote sensing image evaluation model training method, remote sensing image evaluation method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117974571A true CN117974571A (en) | 2024-05-03 |
Family
ID=90857342
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311862159.XA Pending CN117974571A (en) | 2023-12-29 | 2023-12-29 | Remote sensing image evaluation model training method, remote sensing image evaluation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117974571A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112016596A (en) * | 2020-08-10 | 2020-12-01 | 西安科技大学 | Evaluation method for farmland soil fertility based on convolutional neural network |
| CN113496485A (en) * | 2021-06-24 | 2021-10-12 | 北京市遥感信息研究所 | Satellite remote sensing image quality evaluation method and device |
| CN113505854A (en) * | 2021-07-29 | 2021-10-15 | 济南博观智能科技有限公司 | Method, device, equipment and medium for constructing facial image quality evaluation model |
| CN114066831A (en) * | 2021-11-04 | 2022-02-18 | 北京航空航天大学 | Remote sensing image mosaic quality non-reference evaluation method based on two-stage training |
| CN116071623A (en) * | 2023-01-06 | 2023-05-05 | 北京百度网讯科技有限公司 | Model training method, image-based processing method, device, equipment and medium |
| CN116562125A (en) * | 2023-04-07 | 2023-08-08 | 广州市城市规划勘测设计研究院 | Street security simulation prediction method and system using crowd characteristics and street built environment characteristics in coupling mode |
-
2023
- 2023-12-29 CN CN202311862159.XA patent/CN117974571A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112016596A (en) * | 2020-08-10 | 2020-12-01 | 西安科技大学 | Evaluation method for farmland soil fertility based on convolutional neural network |
| CN113496485A (en) * | 2021-06-24 | 2021-10-12 | 北京市遥感信息研究所 | Satellite remote sensing image quality evaluation method and device |
| CN113505854A (en) * | 2021-07-29 | 2021-10-15 | 济南博观智能科技有限公司 | Method, device, equipment and medium for constructing facial image quality evaluation model |
| CN114066831A (en) * | 2021-11-04 | 2022-02-18 | 北京航空航天大学 | Remote sensing image mosaic quality non-reference evaluation method based on two-stage training |
| CN116071623A (en) * | 2023-01-06 | 2023-05-05 | 北京百度网讯科技有限公司 | Model training method, image-based processing method, device, equipment and medium |
| CN116562125A (en) * | 2023-04-07 | 2023-08-08 | 广州市城市规划勘测设计研究院 | Street security simulation prediction method and system using crowd characteristics and street built environment characteristics in coupling mode |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107767413B (en) | Image depth estimation method based on convolutional neural network | |
| CN110046673B (en) | No-reference tone mapping image quality evaluation method based on multi-feature fusion | |
| CN107562805B (en) | Method and device for searching picture by picture | |
| CN104700099B (en) | The method and apparatus for recognizing traffic sign | |
| CN107480649B (en) | Fingerprint sweat pore extraction method based on full convolution neural network | |
| Hosseini et al. | Encoding visual sensitivity by maxpol convolution filters for image sharpness assessment | |
| CN109118445B (en) | Underwater image enhancement method based on multi-branch generation countermeasure network | |
| Po et al. | A novel patch variance biased convolutional neural network for no-reference image quality assessment | |
| CN108399362A (en) | A kind of rapid pedestrian detection method and device | |
| CN110516716B (en) | No-reference image quality evaluation method based on multi-branch similarity network | |
| CN111027590B (en) | Breast cancer data classification method combining deep network features and machine learning model | |
| CN107767358B (en) | Method and device for determining ambiguity of object in image | |
| CN113822982A (en) | Human body three-dimensional model construction method and device, electronic equipment and storage medium | |
| CN110263768A (en) | A kind of face identification method based on depth residual error network | |
| CN109816646B (en) | Non-reference image quality evaluation method based on degradation decision logic | |
| CN109753996B (en) | Hyperspectral image classification method based on three-dimensional lightweight depth network | |
| CN113888501B (en) | Attention positioning network-based reference-free image quality evaluation method | |
| CN109472757A (en) | It is a kind of that logo method is gone based on the image for generating confrontation neural network | |
| CN111862040B (en) | Portrait picture quality evaluation method, device, equipment and storage medium | |
| CN114387270B (en) | Image processing method, image processing device, computer equipment and storage medium | |
| Wei et al. | Sidgan: Single image dehazing without paired supervision | |
| CN111462002A (en) | Underwater image enhancement and restoration method based on convolutional neural network | |
| CN108256557A (en) | The hyperspectral image classification method integrated with reference to deep learning and neighborhood | |
| CN112329818B (en) | Hyperspectral image non-supervision classification method based on graph convolution network embedded characterization | |
| CN111476739B (en) | Underwater image enhancement method, system and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |