CN110147773A - A kind of remote sensing images recognition methods - Google Patents

Abstract

The invention discloses a remote sensing image recognition method. The steps of the method include: acquiring a remote sensing image to be recognized; using a preset remote sensing image recognition model to recognize the remote sensing image to be recognized, and generating a recognition result; wherein, the remote sensing image recognition model is trained by inputting remote sensing image samples into a deep residual network in advance Generated. Compared with the convolutional neural network, the deep residual network skips a certain number of network layers in the process of training remote sensing image samples. Due to the special structure of the deep residual network and the newly introduced idea of residual transfer, The remote sensing image recognition model trained based on the deep residual network can avoid the accuracy degradation caused by the disappearance of the gradient, so that the best accuracy can be preserved, and the recognition accuracy of the remote sensing image is relatively improved. In addition, the present invention also provides a remote sensing image recognition device, equipment, and computer-readable storage medium, the beneficial effects of which are the same as those described above.

Description

A remote sensing image recognition method

technical field

The invention relates to the field of image recognition, in particular to a remote sensing image recognition method.

Background technique

Remote sensing refers to non-contact, long-distance detection technology. Generally refers to the use of sensors / remote sensors to detect the radiation and reflection characteristics of electromagnetic waves of objects. Remote sensing is to detect the target ground objects under the condition of being far away from the target and non-contact target objects through instruments that are sensitive to electromagnetic waves such as remote sensors.

Remote Sensing Image (Remote Sensing Image, referred to as RS) refers to the film or photo that records the electromagnetic wave size of various ground features, mainly divided into aerial photos and satellite photos. Remote sensing images are based on imaging principles and spectral technology, and can capture continuous spectral data and spatial data of target objects. Therefore, hyperspectral images have hundreds of continuous spectral bands and spatial coordinates, and have rich spectral information and spatial information. Spectral information It can be used to accurately identify ground objects, while spatial information can provide supplementary information for spectral features and improve the accuracy of ground object recognition.

Since the types of ground objects contained in remote sensing images are often many and complex, it takes a lot of labor and time to identify the types of ground objects in remote sensing images in an artificial way. The way of network is to identify the types of ground objects in remote sensing images, so as to replace the purpose of manual identification. The continuous increase of the number of network layers will gradually lead to the decline of the network gradient, resulting in unsatisfactory recognition accuracy, and it is difficult to meet the current accuracy requirements for remote sensing image recognition.

It can be seen that providing a remote sensing image recognition method to relatively improve the recognition accuracy of remote sensing images is a problem to be solved by those skilled in the art.

Contents of the invention

The purpose of the present invention is to provide a remote sensing image recognition method to relatively improve the recognition accuracy of the remote sensing image.

In order to solve the above technical problems, the present invention provides a remote sensing image recognition method, comprising:

Obtain remote sensing images to be identified;

The preset remote sensing image recognition model is used to recognize the remote sensing image to be recognized, and the recognition result is generated; wherein, the remote sensing image recognition model is generated by training the remote sensing image sample into the deep residual network in advance.

Preferably, the training of the remote sensing image recognition model includes:

Obtaining remote sensing image samples; wherein, the remote sensing image samples include spectral features, spatial features and color features;

Generate spectral convolution kernel, spatial convolution kernel and color channel convolution kernel based on deep residual network;

Input the remote sensing image samples into the deep residual network, and train the spectral features of the remote sensing image samples through the spectral convolution kernel, train the spatial features of the remote sensing image samples through the spatial convolution kernel, and train the color features of the remote sensing image samples through the color channel convolution kernel , and generate a remote sensing image recognition model.

Preferably, remote sensing image samples are obtained, including:

Obtain remote sensing image samples that have been pre-processed by the PCA algorithm for data dimensionality reduction;

Correspondingly, obtain remote sensing images to be identified, including:

Obtain remote sensing images to be identified that have undergone data dimensionality reduction through the PCA algorithm in advance.

Preferably, before inputting the remote sensing image sample into the deep residual network, the method further includes:

Perform batch normalization on remote sensing image samples.

In addition, the present invention also provides a remote sensing image recognition device, including:

An image acquisition module, configured to acquire remote sensing images to be identified;

The recognition module is configured to use a preset remote sensing image recognition model to recognize the remote sensing image to be recognized and generate a recognition result; wherein, the remote sensing image recognition model is generated by inputting remote sensing image samples into a deep residual network in advance and training.

Preferably, the remote sensing image recognition device also includes:

The sample acquisition module is used to acquire remote sensing image samples; wherein, the remote sensing image samples include spectral features, spatial features and color features;

The convolution kernel generation module is used to generate a spectral convolution kernel, a spatial convolution kernel and a color channel convolution kernel based on a deep residual network;

The sample training module is used to input the remote sensing image samples into the deep residual network, and train the spectral features of the remote sensing image samples through the spectral convolution kernel, train the spatial features of the remote sensing image samples through the spatial convolution kernel, and train the remote sensing image samples through the color channel convolution kernel Color features of remote sensing image samples, and generate remote sensing image recognition models.

In addition, the present invention also provides a remote sensing image recognition device, including:

memory for storing computer programs;

The processor is used to implement the above-mentioned remote sensing image recognition method when executing the computer program.

In addition, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned remote sensing image recognition method is realized.

The remote sensing image recognition method provided by the present invention first obtains the remote sensing image to be recognized, and then uses the preset remote sensing image recognition model to recognize the remote sensing image to be recognized, and generates a recognition result, wherein the remote sensing image recognition model is formed by the depth residual The difference network is generated after training the remote sensing image samples. Compared with the convolutional neural network, the deep residual network skips a certain number of network layers in the process of training remote sensing image samples. Due to the special structure of the deep residual network and the newly introduced idea of residual transfer, The remote sensing image recognition model trained based on the deep residual network can avoid the accuracy degradation caused by the disappearance of the gradient, so that the best accuracy can be preserved, and the recognition accuracy of the remote sensing image is relatively improved. In addition, the present invention also provides a remote sensing image recognition device, equipment, and computer-readable storage medium, the beneficial effects of which are the same as those described above.

Description of drawings

In order to illustrate the embodiments of the present invention more clearly, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. As far as people are concerned, other drawings can also be obtained based on these drawings on the premise of not paying creative work.

Fig. 1 is a flowchart of a remote sensing image recognition method provided by an embodiment of the present invention;

Fig. 2 is the flowchart of the training process of the remote sensing image recognition model provided by the embodiment of the present invention;

Fig. 3 is a structural diagram of a remote sensing image recognition device provided by an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Since the types of ground objects contained in remote sensing images are often many and complex, it takes a lot of labor and time to identify the types of ground objects in remote sensing images in an artificial way. The way of network is to identify the types of ground objects in remote sensing images, so as to replace the purpose of manual identification. The continuous increase of the number of network layers will gradually lead to the decline of the network gradient, resulting in unsatisfactory recognition accuracy, and it is difficult to meet the current accuracy requirements for remote sensing image recognition.

The core of the present invention is to provide a remote sensing image recognition method to relatively improve the recognition accuracy of the remote sensing image. Another core of the present invention is to provide a remote sensing image recognition device, equipment and medium.

In order to enable those skilled in the art to better understand the solution of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

FIG. 1 is a flowchart of a remote sensing image recognition method provided by an embodiment of the present invention. Please refer to Figure 1, the specific steps of the remote sensing image recognition method include:

Step S10: Obtain the remote sensing image to be identified.

It should be noted that the remote sensing image to be identified obtained in this step may be a remote sensing image acquired in real time in a real remote sensing scene, and the essential purpose of this method is to identify the content of the remote sensing image to be identified.

Step S11: Using a preset remote sensing image recognition model to recognize the remote sensing image to be recognized, and generate a recognition result.

Among them, the remote sensing image recognition model is generated by pre-training the remote sensing image samples into the deep residual network.

The focus of this step is to input the remote sensing image samples into the deep residual network in advance for training to generate a remote sensing image recognition model, and then use the remote sensing image recognition model to recognize the acquired remote sensing images to be recognized to generate recognition results. The deep residual network is easier to optimize, and can increase the accuracy by increasing the depth. The core is to solve the side effect of increasing the network depth, that is, the degradation problem. It can improve network performance by simply increasing the network depth. Therefore, when the remote sensing image samples are input to the deep residual network for training, it can relatively avoid the problem that the accuracy of feature extraction in the remote sensing image samples is reduced due to the increase of network depth, thereby ensuring that the remote sensing images generated by the deep residual network training The recognition accuracy of the remote sensing image to be recognized by the recognition model, that is, the recognition accuracy.

The remote sensing image recognition method provided by the present invention first obtains the remote sensing image to be recognized, and then uses the preset remote sensing image recognition model to recognize the remote sensing image to be recognized, and generates a recognition result, wherein the remote sensing image recognition model is formed by the depth residual The difference network is generated after training the remote sensing image samples. Compared with the convolutional neural network, the deep residual network skips a certain number of network layers in the process of training remote sensing image samples. Due to the special structure of the deep residual network and the newly introduced idea of residual transfer, The remote sensing image recognition model trained based on the deep residual network can avoid the accuracy degradation caused by the disappearance of the gradient, so that the best accuracy can be preserved, and the recognition accuracy of the remote sensing image is relatively improved.

On the basis of the above embodiments, the present invention also provides the following series of preferred implementation manners.

Fig. 2 is a flow chart of the training process of the remote sensing image recognition model provided by the embodiment of the present invention. Please refer to Figure 2, the specific steps of training the remote sensing image recognition model include:

Step S20: Acquiring remote sensing image samples.

Among them, remote sensing image samples contain spectral features, spatial features and color features.

It should be noted that the purpose of this step is to obtain remote sensing image samples for training the remote sensing image recognition model. In addition, the remote sensing image samples include spectral features, spatial features and color features.

Among them, the spectral characteristics refer to the characteristics of the pattern of the dispersed monochromatic light arranged in sequence according to the wavelength (or frequency) after the polychromatic light carried in the remote sensing image is split by a dispersion system (such as a prism or grating). The full name is the optical spectrum feature.

Spatial features are the most basic features of geographical phenomena. According to the spatial distribution of geographical phenomena, we can use different spatial dimensions to express them. The distribution of various geographical phenomena has its own inherent spatial distribution characteristics. According to the spatial dimension of geographical phenomena, geographical phenomena can be divided into point distribution, linear distribution, surface distribution and volume distribution, and their spatial relationship is also part of the spatial characteristics.

Color feature is the most widely used visual feature in image retrieval, the main reason is that color is often very related to the objects or scenes contained in the image. In addition, compared with other visual features, color features are less dependent on the size, orientation, and viewing angle of the image itself, and thus have higher robustness. Color features can be subdivided into RGB color features, HIS color features, HSV color features and CMYK color features. In multimedia computer technology, RGB color space representation is the most used. According to the principle of three primary colors, it is represented by primary color light units. The amount of light, in the RGB color space, any color light can be mixed by the addition and mixing of different components of R, G, and B. Compared with other types of color features, the physical meaning is clearer, so it is relatively more accurate. Suitable for training deep residual networks.

Step S21: Generate a spectral convolution kernel, a spatial convolution kernel, and a color channel convolution kernel based on the deep residual network.

The key point of this step is that the spectral convolution kernel for feature training on the spectral features of remote sensing image samples and the spatial convolution kernel for feature training on the spatial features of remote sensing image samples are pre-set in the deep residual network. And the color channel convolution kernel used for feature training on the color features of the remote sensing image samples, and then extract and train the features of the corresponding aspects in the remote sensing image samples through the above three types of convolution kernels.

Step S22: Input the remote sensing image sample into the deep residual network, and train the spectral characteristics of the remote sensing image sample through the spectral convolution kernel, train the spatial characteristics of the remote sensing image sample through the spatial convolution kernel, and train the remote sensing image sample through the color channel convolution kernel color features, and generate a remote sensing image recognition model.

It should be noted that this step is equivalent to inputting the remote sensing image samples into the deep residual network, respectively convolving the three convolution kernels in the deep residual network, namely the spectral convolution kernel, the spatial convolution kernel and the color channel. Kernel, separately train the spectral, spatial and color characteristics of remote sensing image samples, and then generate a remote sensing image recognition model that includes the convolution kernel of the above three aspects of features. Since the training process of the deep residual network based on the data samples belongs to the technical content known to those skilled in the art, it will not be repeated here. In this embodiment, three separate convolution kernels are used to separately train the spectral, spatial, and color features of remote sensing image samples. Since the direction of the trained features is refined, it can be relatively guaranteed that each remote sensing image sample The richness of features during aspect feature extraction improves the training level of detailed features to a certain extent, thereby further ensuring the recognition accuracy of remote sensing image recognition models for remote sensing images.

On the basis of the above embodiments, as a preferred implementation manner, obtaining remote sensing image samples includes:

Obtain remote sensing image samples that have been pre-processed by the PCA algorithm for data dimensionality reduction;

Correspondingly, obtain remote sensing images to be identified, including:

Obtain remote sensing images to be identified that have undergone data dimensionality reduction through the PCA algorithm in advance.

It should be noted that the PCA (Principal components analysis) algorithm, that is, the principal component analysis algorithm, is mainly used to reduce the dimension of the data set, while retaining the features that contribute the most to the variance in the data, retaining low-order components and removing high-order components. The low-order components tend to contain the most important aspects of the data, while making the remaining components orthogonal to each other.

In this embodiment, before training the remote sensing image recognition model, data dimensionality reduction is performed on the remote sensing image samples through the PCA algorithm in advance, and before the remote sensing image to be recognized is recognized through the remote sensing image recognition model, the remote sensing image to be recognized is also preliminarily passed through the PCA algorithm The main purpose of data dimensionality reduction is to reduce the spectral dimension of remote sensing images and remove some unimportant spectral data. Since each spectrum is orthogonal to each other, it can reduce the factors that affect each other between the original remote sensing data spectra and achieve compressed input. The purpose of the data is to make the remote sensing image samples and the data of the remote sensing images to be recognized better convoluted, further improve the training degree of the deep residual network for the remote sensing image samples, and further reduce the remote sensing image recognition model for the recognition of the remote sensing images to be recognized. In order to improve the recognition accuracy of the remote sensing image recognition model for remote sensing images.

On the basis of the above series of embodiments, as a preferred implementation, before inputting the remote sensing image samples into the deep residual network, the method further includes:

Perform batch normalization on remote sensing image samples.

It should be noted that batch normalization processing of remote sensing image samples before inputting remote sensing image samples into the deep residual network can speed up the convergence speed of the deep residual network after the remote sensing image samples are input into the deep residual network. It can solve the gradient dispersion problem of the deep residual network, that is, when the gradient descent method is used, the weights of the first few layers change very slowly, so that they cannot effectively learn from the samples. This embodiment makes the deep residual network The difference network is more stable during training, ensuring the overall reliability of the generated remote sensing image recognition model.

Fig. 3 is a structural diagram of a remote sensing image recognition device provided by an embodiment of the present invention. The remote sensing image recognition device provided by the embodiment of the present invention includes:

The image acquisition module 10 is configured to acquire remote sensing images to be identified.

The recognition module 11 is configured to use a preset remote sensing image recognition model to recognize the remote sensing image to be recognized and generate a recognition result; wherein, the remote sensing image recognition model is generated by inputting remote sensing image samples into a deep residual network in advance and training.

The remote sensing image recognition device provided by the present invention first obtains the remote sensing image to be recognized, and then uses the preset remote sensing image recognition model to recognize the remote sensing image to be recognized, and generates a recognition result, wherein the remote sensing image recognition model is formed by the depth residual The difference network is generated after training the remote sensing image samples. Compared with the convolutional neural network, the deep residual network skips a certain number of network layers in the process of training remote sensing image samples. Due to the special structure of the deep residual network and the newly introduced idea of residual transfer, The remote sensing image recognition model trained based on the deep residual network can avoid the accuracy degradation caused by the disappearance of the gradient, so that the best accuracy can be preserved, and the recognition accuracy of the remote sensing image is relatively improved.

The present invention also provides a remote sensing image recognition device, comprising:

memory for storing computer programs;

The processor is configured to implement the remote sensing image recognition method as described above when executing the computer program.

The remote sensing image recognition device provided by the present invention first acquires the remote sensing image to be recognized, and then uses the preset remote sensing image recognition model to recognize the remote sensing image to be recognized, and generates a recognition result, wherein the remote sensing image recognition model is formed by the depth residual The difference network is generated after training the remote sensing image samples. Compared with the convolutional neural network, the deep residual network skips a certain number of network layers in the process of training remote sensing image samples. Due to the special structure of the deep residual network and the newly introduced idea of residual transfer, The remote sensing image recognition model trained based on the deep residual network can avoid the accuracy degradation caused by the disappearance of the gradient, so that the best accuracy can be preserved, and the recognition accuracy of the remote sensing image is relatively improved.

The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned remote sensing image recognition method are realized.

The computer-readable storage medium provided by the present invention first acquires the remote sensing image to be recognized, and then uses the preset remote sensing image recognition model to recognize the remote sensing image to be recognized, and generates a recognition result, wherein the remote sensing image recognition model is determined by the depth The residual network is generated after training the remote sensing image samples. Compared with the convolutional neural network, the deep residual network skips a certain number of network layers in the process of training remote sensing image samples. Due to the special structure of the deep residual network and the newly introduced idea of residual transfer, The remote sensing image recognition model trained based on the deep residual network can avoid the accuracy degradation caused by the disappearance of the gradient, so that the best accuracy can be preserved, and the recognition accuracy of the remote sensing image is relatively improved.

A remote sensing image recognition method provided by the present invention has been introduced in detail above. Each embodiment in the description is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other. As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for the related information, please refer to the description of the method part. It should be pointed out that for those skilled in the art, without departing from the principles of the present invention, some improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

It should also be noted that in this specification, relative terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that these entities or operations There is no such actual relationship or order between the operations. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

Claims (7)

1. A remote sensing image recognition method, characterized in that, comprising: Obtain remote sensing images to be identified; The remote sensing image to be recognized is recognized by using a preset remote sensing image recognition model to generate a recognition result; wherein the remote sensing image recognition model is generated by inputting remote sensing image samples into a deep residual network in advance and training. 2. remote sensing image recognition method according to claim 1, is characterized in that, the training of described remote sensing image recognition model comprises: Acquiring the remote sensing image sample; wherein, the remote sensing image sample includes spectral features, spatial features and color features; Generate a spectral convolution kernel, a spatial convolution kernel, and a color channel convolution kernel based on the deep residual network; inputting the remote sensing image samples into the deep residual network, and training the spectral features of the remote sensing image samples through the spectral convolution kernel, training the spatial features of the remote sensing image samples through the spatial convolution kernel, and The color channel convolution kernel trains the color features of the remote sensing image samples, and generates the remote sensing image recognition model. 3. The remote sensing image recognition method according to claim 2, wherein said obtaining said remote sensing image sample comprises: Obtaining the remote sensing image samples that have been subjected to data dimensionality reduction through the PCA algorithm in advance; Correspondingly, the acquisition of remote sensing images to be identified includes: Obtain the remote sensing image to be identified that has undergone data dimensionality reduction through the PCA algorithm in advance. 4. The remote sensing image recognition method according to claim 2 or 3, wherein, before the remote sensing image sample is input into the deep residual network, the method further comprises: Perform batch normalization processing on the remote sensing image samples. 5. A remote sensing image recognition device, characterized in that it comprises: An image acquisition module, configured to acquire remote sensing images to be identified; The recognition module is configured to use a preset remote sensing image recognition model to recognize the remote sensing image to be recognized and generate a recognition result; wherein, the remote sensing image recognition model is generated by training the remote sensing image sample into the deep residual network in advance . 6. A remote sensing image recognition device, characterized in that it comprises: memory for storing computer programs; A processor configured to implement the remote sensing image recognition method according to any one of claims 1 to 4 when executing the computer program. 7. A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the remote sensing as described in any one of claims 1 to 4 is realized image recognition method.