CN114648706B - Forest tree species identification method, device and equipment based on satellite remote sensing image - Google Patents
Forest tree species identification method, device and equipment based on satellite remote sensing image Download PDFInfo
- Publication number
- CN114648706B CN114648706B CN202210561378.3A CN202210561378A CN114648706B CN 114648706 B CN114648706 B CN 114648706B CN 202210561378 A CN202210561378 A CN 202210561378A CN 114648706 B CN114648706 B CN 114648706B
- Authority
- CN
- China
- Prior art keywords
- image
- remote sensing
- tree species
- network
- satellite remote
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/251—Fusion techniques of input or preprocessed data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The application discloses a forest tree species identification method, a forest tree species identification device and forest tree species identification equipment based on a satellite remote sensing image, wherein different spectral information in the satellite remote sensing image is effectively and fully fused by generating a multi-channel image, so that subsequent tree species identification is more accurate; increasing equivalent depth through a pre-trained residual error network to improve the accuracy of subsequent tree species identification; through the counterstudy of the generator and the discriminator, the neural network can extract higher features, the fine granularity is improved, and the accuracy of tree species identification and classification is further improved.
Description
Technical Field
The application relates to the field of satellite remote sensing, in particular to a forest tree species identification method, device and equipment based on a satellite remote sensing image.
Background
The forest is the most important land ecosystem on the earth, the tree species distribution in the forest is an important index for monitoring forest resources, the composition and the distribution of the forest tree species are closely related to factors such as forest biomass, biodiversity, forest quality and the like, and in addition, forest fire prevention, forest pest and disease estimation, forest change information extraction and the like all depend on high-precision forest tree species identification. The method for analyzing the tree species distribution by using the implementation, the dynamic and the comprehensive of the remote sensing technology is the most widely used means with the highest monitoring efficiency at present.
The tree species identification technology based on the remote sensing technology is used for classifying tree species through spectrum and space information provided by a high-resolution image acquired by a remote sensing platform, such as methods of a vector machine (SVM), a decision tree and the like, but the identification accuracy of the existing tree species identification method is low.
Disclosure of Invention
The method, the device and the equipment for identifying the forest tree species based on the satellite remote sensing image aim at solving the technical problem that an existing tree species identification method is low in identification accuracy.
In order to achieve the above object, the present application provides a forest tree species identification method based on a satellite remote sensing image, including:
obtaining a satellite remote sensing image, wherein the satellite remote sensing image comprises a plurality of wave bands;
carrying out wave band fusion on the satellite remote sensing image to obtain a multi-channel image;
obtaining an image training set and a label image according to the multi-channel image;
obtaining a pre-training weight according to the image training set and a pre-training residual error network;
obtaining a prediction result according to the pre-training weight and the generator network;
and obtaining a tree species identification result according to the prediction result, the label image and a discriminator network.
Optionally, the step of performing band fusion on the satellite remote sensing image to obtain a multi-channel image includes:
fusing each wave band of the satellite remote sensing image to obtain a true color image and a normalized vegetation index;
and obtaining the multichannel image according to the true color image and the normalized vegetation index.
Optionally, the step of obtaining a prediction result according to the pre-training weights and the generator network includes:
training the generator network to obtain the prediction by the following relation:
wherein L is seg Is the generator loss function, L ce Is a semantic segmentation loss function, L adv Is a function of the penalty of fighting, L semi Is a semi-supervised loss function, λ adv And λ semi Are each L adv And L semi The weight of (c).
Optionally, the expression of the semi-supervised loss function is:
wherein L is semi Is a semi-supervised loss function, S is a generator network, D is a discriminator network, c is a tree species category, (h, w) are position coordinates, T semi Is a threshold value for controlling the sensitivity of the self-learning process, and I is an index function of the image training set.
Optionally, the expression of the penalty function is:
wherein L is adv Is the penalty function, S is the generator network, D is the discriminator network, and (h, w) are the position coordinates.
Optionally, the expression of the semantic segmentation loss function is:
wherein L is ce Is the semantic segmentation loss function, S is the generator network, c is the tree species class, and (h, w) is the location coordinates.
Optionally, the step of obtaining a tree species identification result according to the prediction result, the label image and a discriminator network includes:
training the discriminator network to obtain the tree species recognition result according to the following relation:
wherein L is D Is the discriminator loss function, S is the generator network, D is the discriminator network, D (X) n )) (h,w) Is a confidence map of the input image X at the position coordinates (h, w).
Optionally, the last layer of the discriminator network is an upsampling layer, and is configured to output the tree species identification result as an image with the same format as the multi-channel image.
In addition, in order to realize above-mentioned purpose, this application still provides a forest tree kind recognition device based on satellite remote sensing image, includes:
the satellite remote sensing image acquisition module is used for acquiring a satellite remote sensing image, and the satellite remote sensing image comprises a plurality of wave bands;
the band fusion module is used for carrying out band fusion on the satellite remote sensing image to obtain a multi-channel image;
the image training set and label image acquisition module is used for acquiring an image training set and a label image according to the multi-channel image;
the weight acquisition module is used for acquiring pre-training weights according to the image training set and a pre-training residual error network;
the prediction result acquisition module is used for acquiring a prediction result according to the pre-training weight and the generator network;
and the tree species identification result acquisition module is used for acquiring a tree species identification result according to the prediction result, the label image and the discriminator network.
In addition, to achieve the above object, the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above method.
The beneficial effect that this application can realize.
According to the forest tree species identification method, device and equipment based on the satellite remote sensing image, the satellite remote sensing image is obtained, and the satellite remote sensing image comprises a plurality of wave bands; carrying out wave band fusion on the satellite remote sensing image to obtain a multi-channel image; obtaining an image training set and a label image according to the multi-channel image; obtaining a pre-training weight according to the image training set and a pre-training residual error network; obtaining a prediction result according to the pre-training weight and the generator network; and obtaining a tree species identification result according to the prediction result, the label image and a discriminator network. By generating a multi-channel image, different spectral information in the satellite remote sensing image is effectively and fully fused, so that the subsequent tree species identification is more accurate; increasing equivalent depth through a pre-trained residual error network to improve the accuracy of subsequent tree species identification; through the counterstudy of the generator and the discriminator, the neural network can extract higher features, the fine granularity is improved, and the accuracy of tree species identification and classification is further improved.
Drawings
FIG. 1 is a schematic diagram of a computer device in a hardware operating environment according to an embodiment of the present application;
fig. 2 is a schematic flow chart of a forest tree species identification method based on a satellite remote sensing image according to an embodiment of the present application;
fig. 3 is a functional module schematic diagram of a forest tree species recognition device based on a satellite remote sensing image according to an embodiment of the present application.
The implementation, functional features and advantages of the objectives of the present application will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
The main solution of the embodiment of the application is as follows: according to the forest tree species identification method, device and equipment based on the satellite remote sensing image, the satellite remote sensing image is obtained, and the satellite remote sensing image comprises a plurality of wave bands; carrying out wave band fusion on the satellite remote sensing image to obtain a multi-channel image; obtaining an image training set and a label image according to the multi-channel image; obtaining a pre-training weight according to the image training set and a pre-training residual error network; obtaining a prediction result according to the pre-training weight and the generator network; and obtaining a tree species identification result according to the prediction result, the label image and a discriminator network.
In the prior art, a forest is the most important land ecosystem on the earth, the tree species distribution in the forest is an important index for monitoring forest resources, the composition and the distribution of the forest tree species are closely related to factors such as forest biomass, biodiversity and forest quality, and in addition, forest fire prevention, forest pest and disease damage estimation, forest change information extraction and the like all depend on high-precision forest tree species identification. The method for analyzing the tree species distribution by using the implementation, the dynamic and the comprehensive of the remote sensing technology is the most widely used means with the highest monitoring efficiency at present.
The tree species identification technology based on the remote sensing technology is used for classifying tree species through spectrum and space information provided by a high-resolution image acquired by a remote sensing platform, such as methods of a vector machine (SVM), a decision tree and the like, but the identification accuracy of the existing tree species identification method is low. Meanwhile, the traditional tree species classification method needs a large amount of label sample data, and the manual marking has the disadvantages of large workload, long time and high economic cost.
Therefore, the method and the device provide a solution, different spectral information in the satellite remote sensing image is effectively fused by generating the multi-channel image, the forest characteristics are more prominent by normalizing the vegetation index, and the accuracy of subsequent tree species identification is improved; by generating the loss of the countermeasure network, the neural network is guided to extract higher features so as to help the semantic segmentation of fine granularity, so that the tree species identification information is more accurate; by applying the semi-supervised learning method, only a small number of labels are needed in the tree species identification process, the labeling quantity of semantic segmentation is reduced, and further, a large amount of time and economic cost are reduced.
Referring to fig. 1, fig. 1 is a schematic structural diagram of a computer device in a hardware operating environment according to an embodiment of the present application.
As shown in fig. 1, the computer apparatus may include: a processor 1001, such as a Central Processing Unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a WIreless interface (e.g., a WIreless-FIdelity (WI-FI) interface). The Memory 1005 may be a Random Access Memory (RAM) Memory, or may be a Non-Volatile Memory (NVM), such as a disk Memory. The memory 1005 may alternatively be a storage device separate from the processor 1001.
Those skilled in the art will appreciate that the configuration shown in FIG. 1 does not constitute a limitation of a computer device and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components.
As shown in fig. 1, a memory 1005, which is a storage medium, may include therein an operating system, a data storage module, a network communication module, a user interface module, and an electronic program.
In the computer device shown in fig. 1, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 of the computer device may be arranged in the computer device, and the computer device calls the forest tree species identification device based on the satellite remote sensing image stored in the memory 1005 through the processor 1001 and executes the forest tree species identification method based on the satellite remote sensing image provided by the embodiment of the present invention.
Referring to fig. 2, based on the hardware device in the foregoing embodiment, an embodiment of the present application provides a forest tree species identification method based on a satellite remote sensing image, including:
s10: obtaining a satellite remote sensing image, wherein the satellite remote sensing image comprises a plurality of wave bands;
in the specific implementation process, satellite remote sensing refers to a general term of various comprehensive technical systems for observing the earth and the celestial body from the ground to the space, and satellite remote sensing images can be acquired from a remote sensing technical platform and information can be processed and analyzed. According to the technical scheme, the remote sensing image is obtained mainly through a high resolution second (GF-2) satellite, the space resolution is better than 1 meter of civil optical remote sensing satellite, and other optical remote sensing satellites meeting the resolution and channel information requirements can be selected.
S20: carrying out wave band fusion on the satellite remote sensing image to obtain a multi-channel image;
in a specific implementation process, the multi-channel image is an image with the number of channels being greater than or equal to 3. The high-resolution second satellite comprises a blue wave band, a green wave band, a red wave band and an infrared wave band, the wave bands are fused to obtain a multi-channel image, and different spectral information can be effectively fused to identify tree species.
As an optional implementation manner, the step of performing band fusion on the satellite remote sensing image to obtain a multi-channel image includes: fusing each wave band of the satellite remote sensing image to obtain a true color image and a normalized vegetation index; and obtaining the multichannel image according to the true color image and the normalized vegetation index.
In the specific implementation process, the blue waveband, the green waveband and the red waveband are fused to obtain a true color image, wherein the true color image is formed by R, G and B primary color components in each pixel value forming a color image, and each primary color component directly determines the primary color intensity of the display device, so that the color characteristics of actual ground objects can be truly reflected.
The normalized vegetation index is calculated by the following relationship:
NDVI=(B4-B3)/(B4+B3)
wherein, B3 refers to red wave band, B4 refers to infrared wave band, NDVI refers to normalized vegetation index, the index is used for inverting ecological characteristics of crops, different crops have different spectral emission characteristics in different growth periods, and different tree species types can be distinguished. And after the NDVI characteristics are converted into binary images, combining the binary images with true color images to form a multi-channel image.
S30: obtaining an image training set and a label image according to the multi-channel image;
in the specific implementation process, the multi-channel image is cut into an image with the length and the width of 512 pixels, and the image has both detection speed and recognition effect. Selecting a small number of image labeling tree species information labels as standardized label images; and taking the residual images as an image training set to facilitate subsequent neural network training, wherein the number of the images in the image training set is far larger than that of the label images. And a small amount of label samples are used for training, so that the marking cost is saved.
S40: obtaining a pre-training weight according to the image training set and a pre-training residual error network;
in the specific implementation process, the residual error network is converted into a corresponding residual error version by inserting a quick link on the basis of a simple network, and the residual error version is not directly fitted with a target but fitted with a residual error; it is easier to optimize and can improve the accuracy of subsequent tree species identification by increasing the comparable depth. In this embodiment, a residual network resnet101 is pre-trained on Imagnet, and the resnet101 network includes 4 volume blocks, a pooling layer and a full connection layer.
Inputting the image training set number into a resnet101 network to obtain a pre-training weight, wherein the input of the pre-training weight is three-channel, only the predefined three-channel weight value is assigned to an RGB channel by modifying the weight of a predefined model convolution layer 1, and the NDVI channel is initialized by using Keming.
S50: obtaining a prediction result according to the pre-training weight and the generator network;
in a specific implementation process, the pre-training weights are trained through a generator network, each pixel is classified according to the pre-training weights, and species identification is carried out, so that a prediction result is obtained. In this embodiment, a DeepLab-v3+ network is used as a generator network, and a prediction result is obtained by the following relational expression:
P=S（X n ，W b ）
where Xn represents the input multi-channel image and S represents a parameter W b DeepLab-v3+ network.
As an alternative embodiment, the step of obtaining the prediction result according to the pre-training weights and the generator network includes: training the generator network to obtain the prediction by the following relation:
wherein L is seg Is the generator loss function, L ce Is a semantic segmentation loss function, L adv Is a function of the penalty of fighting, L semi Is a semi-supervised loss function, λ adv And λ semi Are each L adv And L semi The weight of (c).
In a specific implementation, loss function is constructed by constructing a generatorL seg The generator network is trained, and joint optimization is carried out by minimizing a plurality of losses such as semantic segmentation, semi-supervision, counterlearning and the like. Lambda [ alpha ] adv And λ semi These two weights are used to minimize the multitask penalty function.
As an optional implementation, the expression of the semi-supervised loss function is:
wherein L is semi Is a semi-supervised loss function, S is a generator network, D is a discriminator network, c is a tree species category, (h, w) are position coordinates, T semi Is a threshold value for controlling the sensitivity of the self-learning process, and I is an index function of the image training set.
In the specific implementation process, a Semi-Supervised Learning (SSL) method is a Learning method combining Supervised Learning and unsupervised Learning, and a large amount of unmarked data and a small amount of marked data are used for pattern recognition, so that time and economic cost brought by manual marking can be greatly saved, and higher accuracy can be ensured. And a small amount of label images are used for training the network, so that the information of the image training set is more accurate.
As an alternative embodiment, the expression of the penalty function is:
wherein L is adv Is the penalty function, S is the generator network, D is the discriminator network, and (h, w) are the position coordinates.
In the specific implementation process, the countermeasure refers to a countermeasure generation network (GAN), which mainly includes two parts, namely a generator and a discriminator, wherein the generator is mainly used for learning real image distribution so as to make the self-generated image more real, and thus cheat the discriminator, and the discriminator needs to reject the received imageAnd (5) judging whether the picture is true or false. In the whole process, the generator enables the generated image to be more real, the discriminator identifies the truth of the image, the process is equivalent to mutual game, and finally two networks reach dynamic balance along with the continuous counterwork of the generator and the discriminator: the image generated by the generator is close to the real image distribution, and the discriminator can not identify the real and false images. This embodiment works by countering the loss function L adv And performing an antagonistic learning process, and guiding the neural network to extract higher features by generating the loss of the antagonistic network so as to help the semantic segmentation with fine granularity, so that the information of the image training set is more real, and the tree species identification is more accurate.
As an optional implementation, the expression of the semantic segmentation loss function is:
wherein L is ce Is the semantic segmentation loss function, S is the generator network, c is the tree species class, and (h, w) is the location coordinates.
In a specific implementation, semantic segmentation is to understand an image from the pixel level, i.e., to assign a class label to each pixel in the image. The embodiment uses the deplab-v 3+ semantic segmentation network, is an encoder-Decoder with cavity separable convolution for semantic image segmentation, and is implemented by adding a simple and effective Decoder on the deplab-v 3 to refine segmentation results, especially segmentation results along the boundary of a target object, and by adopting a mode of integrating a spatial pyramid pool module or a coding and decoding structure. Loss function L by semantic segmentation ce And comparing the prediction output of the semantic segmentation network with the label result to optimize the network, and obtaining deep semantic features to extract high-dimensional semantic information and improve the fine-grained classification level of tree species.
S60: and obtaining a tree species identification result according to the prediction result, the label image and a discriminator network.
In the specific implementation process, the discriminator network consists of 5 convolution layers of 4 x 4 and channels of {64,128,256,512 and 1} with the step length of 2, an active layer leak-ReLU with the parameter of 0.2 is arranged behind each convolution layer, the prediction result and the label image are input into the discriminator network together for the counterstudy, the image training set is continuously optimized to be closer to the distribution of the real image, namely closer to the label image, and the tree species recognition result is obtained.
As an optional implementation manner, the step of obtaining a tree species identification result according to the prediction result, the tag image and a discriminator network includes: training the discriminator network to obtain the tree species recognition result according to the following relation:
wherein L is D Is the discriminator loss function, S is the generator network, D is the discriminator network, D (X) n )) (h,w) Is a confidence map of the input image X at the position coordinates (h, w).
In the concrete implementation process, a discriminator network is trained by constructing a discriminator loss function, and the spatial cross entropy is used for losing
Minimization, arbiter output result y n =0 meaning that the image is output from the generator network, y n =0 means that the image is extracted from the label image.
As an optional implementation manner, the last layer of the discriminator network is an upsampling layer, and is used for outputting the tree species identification result as an image with the same format as the multi-channel image.
In particular implementations, upsampling may allow the target image to be changed to a higher resolution. The last layer of the discriminator network in this embodiment is an upsampling layer, and the output tree species recognition result image is scaled to the same format as the input multi-channel image, that is, 512 pixels in size. The output result has the same format as the input image, realizes the automatic identification of the tree species from end to end, and the output identification result does not need to carry out the subsequent format conversion processing.
It should be understood that the above is only an example, and the technical solution of the present application is not limited in any way, and those skilled in the art can make the setting based on the actual application, and the setting is not limited herein.
According to the description, different spectral information in the satellite remote sensing image is effectively fused by generating the multi-channel image, the forest characteristics are more prominent by normalizing the vegetation index, and the accuracy of subsequent tree species identification is improved; by generating the loss of the countermeasure network, the neural network is guided to extract higher features so as to help the semantic segmentation of fine granularity, so that the tree species identification information is more accurate; by applying the semi-supervised learning method, only a small number of labels are needed in the tree species identification process, the labeling quantity of semantic segmentation is reduced, and further, a large amount of time and economic cost are reduced.
Referring to fig. 3, based on the same inventive concept, an embodiment of the present application further provides a forest tree species identification device based on a satellite remote sensing image, including:
the satellite remote sensing image acquisition module is used for acquiring a satellite remote sensing image, and the satellite remote sensing image comprises a plurality of wave bands;
the band fusion module is used for carrying out band fusion on the satellite remote sensing image to obtain a multi-channel image;
the image training set and label image acquisition module is used for acquiring an image training set and a label image according to the multi-channel image;
the weight acquisition module is used for acquiring pre-training weights according to the image training set and a pre-training residual error network;
the prediction result acquisition module is used for acquiring a prediction result according to the pre-training weight and the generator network;
and the tree species identification result acquisition module is used for acquiring a tree species identification result according to the prediction result, the label image and the discriminator network.
It should be noted that, in this embodiment, each module in the forest tree species identification device based on the satellite remote sensing image corresponds to each step in the forest tree species identification method based on the satellite remote sensing image in the foregoing embodiment one to one, and therefore, a specific implementation of this embodiment may refer to the implementation of the forest tree species identification method based on the satellite remote sensing image, and details are not described here.
Furthermore, in an embodiment, an embodiment of the present application further provides a computer device, which includes a processor, a memory, and a computer program stored in the memory, and when the computer program is executed by the processor, the steps of the method in the foregoing embodiments are implemented.
In some embodiments, the memory may be FRAM, ROM, PROM, EPROM, EEPROM, flash, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories. The computer may be a variety of computing devices including intelligent terminals and servers.
In some embodiments, the executable instructions may be in the form of a program, software module, script, or code written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and it may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.
By way of example, executable instructions may correspond, but do not necessarily have to correspond, to files in a file system, and may be stored in a portion of a file that holds other programs or data, such as in one or more scripts in a hypertext Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code).
By way of example, executable instructions may be deployed to be executed on one computing device or on multiple computing devices at one site or distributed across multiple sites and interconnected by a communication network.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The above-mentioned serial numbers of the embodiments of the present application are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present application may be embodied in the form of a software product, which is stored in a storage medium (e.g., a rom/ram, a magnetic disk, an optical disk) and includes instructions for enabling a multimedia terminal (e.g., a mobile phone, a computer, a television receiver, or a network device) to execute the method according to the embodiments of the present application.
The above description is only a preferred embodiment of the present application, and not intended to limit the scope of the present application, and all modifications of equivalent structures and equivalent processes, which are made by the contents of the specification and the drawings of the present application, or which are directly or indirectly applied to other related technical fields, are included in the scope of the present application.
Claims (5)
1. A forest tree species identification method based on satellite remote sensing images is characterized by comprising the following steps:
obtaining a satellite remote sensing image, wherein the satellite remote sensing image comprises a plurality of wave bands;
carrying out wave band fusion on the satellite remote sensing image to obtain a multi-channel image;
obtaining an image training set and a label image according to the multi-channel image;
obtaining a pre-training weight according to the image training set and a pre-training residual error network;
training a generator network through the following relational expression, and obtaining a prediction result according to the pre-training weight and the generator network; wherein L is seg Is the generator loss function, L ce Is a semantic segmentation loss function, L adv Is a function of the penalty semi Is a semi-supervised loss function, λ adv And λ semi Are each L adv And L semi S is a generator network, D is a discriminator network, c is a tree species category, (h, w) is a position coordinate, T semi Is a threshold value for controlling the sensitivity of the self-learning process, and I is an index function of an image training set;
training the discriminator network according to the following relation, and according to the prediction result, the label image and the discriminator networkObtaining tree species identification results; wherein L is D Is a discriminant loss function, S is a generator network, D is a discriminant network, D (S (xn)) (h, w) is a confidence map of the input image X at the position coordinates (h, w);
2. the forest tree species recognition method based on the satellite remote sensing image as claimed in claim 1, wherein the step of performing band fusion on the satellite remote sensing image to obtain a multi-channel image comprises the steps of:
fusing each wave band of the satellite remote sensing image to obtain a true color image and a normalized vegetation index;
and obtaining the multichannel image according to the true color image and the normalized vegetation index.
3. A forest tree species recognition method based on satellite remote sensing images as claimed in claim 1, characterized in that the last layer of the discriminator network is an upsampling layer for outputting the tree species recognition results as images with the same format as the multi-channel images.
4. The utility model provides a forest tree species recognition device based on satellite remote sensing image which characterized in that includes:
the satellite remote sensing image acquisition module is used for acquiring a satellite remote sensing image, and the satellite remote sensing image comprises a plurality of wave bands;
the band fusion module is used for carrying out band fusion on the satellite remote sensing image to obtain a multi-channel image;
the image training set and label image acquisition module is used for acquiring an image training set and a label image according to the multi-channel image;
the weight acquisition module is used for acquiring pre-training weights according to the image training set and a pre-training residual error network;
prediction result acquisition moduleThe generator network is trained through the following relational expression, and a prediction result is obtained according to the pre-training weight and the generator network; wherein L is seg Is the generator loss function, L ce Is a semantic segmentation loss function, L adv Is a function of the penalty of fighting, L semi Is a semi-supervised loss function, λ adv And λ semi Are each L adv And L semi S is a generator network, D is a discriminator network, c is a tree species category, (h, w) is a position coordinate, T semi Is a threshold value for controlling the sensitivity of the self-learning process, and I is an index function of an image training set;
the tree species identification result acquisition module is used for training the discriminator network through the following relational expression and acquiring a tree species identification result according to the prediction result, the label image and the discriminator network; wherein L is D Is a discriminant loss function, S is a generator network, D is a discriminant network, D (S (xn)) (h, w) is a confidence map of the input image X at the position coordinates (h, w);
5. a computer device, characterized in that it comprises a memory in which a computer program is stored and a processor which executes said computer program implementing the method according to any one of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210561378.3A CN114648706B (en) | 2022-05-23 | 2022-05-23 | Forest tree species identification method, device and equipment based on satellite remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210561378.3A CN114648706B (en) | 2022-05-23 | 2022-05-23 | Forest tree species identification method, device and equipment based on satellite remote sensing image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114648706A CN114648706A (en) | 2022-06-21 |
| CN114648706B true CN114648706B (en) | 2022-07-29 |
Family
ID=81997385
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210561378.3A Active CN114648706B (en) | 2022-05-23 | 2022-05-23 | Forest tree species identification method, device and equipment based on satellite remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114648706B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631401A (en) * | 2015-12-11 | 2016-06-01 | 中国烟草总公司广东省公司 | Remote sensing and geographic information-based tobacco field tobacco quality monitoring and prediction method |
| CN109815914A (en) * | 2019-01-28 | 2019-05-28 | 成都蝉远科技有限公司 | A kind of convolutional neural networks model training method and system based on vegetation area identification |
| CN112132149A (en) * | 2020-09-10 | 2020-12-25 | 武汉汉达瑞科技有限公司 | Semantic segmentation method and device for remote sensing image |
| CN112580504A (en) * | 2020-12-17 | 2021-03-30 | 中国科学院空天信息创新研究院 | Tree species classification counting method and device based on high-resolution satellite remote sensing image |
| CN112633142A (en) * | 2020-12-21 | 2021-04-09 | 广东电网有限责任公司电力科学研究院 | Power transmission line violation building identification method and related device |
| CN113468951A (en) * | 2021-05-20 | 2021-10-01 | 华东师范大学 | Method for detecting mangrove landform based on Landsat satellite remote sensing |
| CN113591766A (en) * | 2021-08-09 | 2021-11-02 | 中国林业科学研究院资源信息研究所 | Multi-source remote sensing tree species identification method for unmanned aerial vehicle |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110687049B (en) * | 2018-07-04 | 2022-12-20 | 中国科学院深圳先进技术研究院 | Method and device for identifying timber forest |
| CN109522807B (en) * | 2018-10-22 | 2020-11-27 | 深圳先进技术研究院 | Satellite image recognition system and method based on self-generated features and electronic equipment |
| CN115668307A (en) * | 2020-05-22 | 2023-01-31 | 巴斯夫欧洲公司 | System and method for estimating vegetation coverage in a real world environment |
| CN113609889B (en) * | 2021-05-18 | 2024-02-13 | 武汉大学 | High-resolution remote sensing image vegetation extraction method based on sensitive characteristic focusing perception |
-
2022
- 2022-05-23 CN CN202210561378.3A patent/CN114648706B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631401A (en) * | 2015-12-11 | 2016-06-01 | 中国烟草总公司广东省公司 | Remote sensing and geographic information-based tobacco field tobacco quality monitoring and prediction method |
| CN109815914A (en) * | 2019-01-28 | 2019-05-28 | 成都蝉远科技有限公司 | A kind of convolutional neural networks model training method and system based on vegetation area identification |
| CN112132149A (en) * | 2020-09-10 | 2020-12-25 | 武汉汉达瑞科技有限公司 | Semantic segmentation method and device for remote sensing image |
| CN112580504A (en) * | 2020-12-17 | 2021-03-30 | 中国科学院空天信息创新研究院 | Tree species classification counting method and device based on high-resolution satellite remote sensing image |
| CN112633142A (en) * | 2020-12-21 | 2021-04-09 | 广东电网有限责任公司电力科学研究院 | Power transmission line violation building identification method and related device |
| CN113468951A (en) * | 2021-05-20 | 2021-10-01 | 华东师范大学 | Method for detecting mangrove landform based on Landsat satellite remote sensing |
| CN113591766A (en) * | 2021-08-09 | 2021-11-02 | 中国林业科学研究院资源信息研究所 | Multi-source remote sensing tree species identification method for unmanned aerial vehicle |
Non-Patent Citations (6)
| Title |
|---|
| Assessment of Classifiers and Remote Sensing Features of Hyperspectral Imagery and Stereo-Photogrammetric Point Clouds for Recognition of Tree Species in a Forest Area of High Species Diversity;Sakari Tuominen等;《remote sensing》;20180505;第1-28页 * |
| Semi-Supervised Semantic Segmentation with High- and Low-level Consistency;Sudhanshu Mittal等;《IEEE Transactions on Pattern Analysis and Machine Intelligence》;20190815;第1-11页 * |
| 基于多特征融合和 CNN 模型的树种图像识别研究;刘嘉政 等;《北京林业大学学报》;20191130;第41卷(第11期);第76-86页 * |
| 基于改进U-Net的高光谱农林植被分类方法;王克奇 等;《森林工程》;20220131;第38卷(第1期);第58-66页 * |
| 基于深度语义分割的无人机影像烟草种植面积提取;付必环 等;《通信技术》;20220228;第55卷(第2期);第181-186页 * |
| 基于高光谱遥感数据的森林树种分类关键技术研究;赵磊;《中国优秀硕士学位论文全文数据库 农业科技辑》;20220515;第D049-31页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114648706A (en) | 2022-06-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sadeghi-Tehran et al. | DeepCount: in-field automatic quantification of wheat spikes using simple linear iterative clustering and deep convolutional neural networks | |
| CN107767408B (en) | Image processing method, processing device and processing equipment | |
| CN111126258A (en) | Image recognition method and related device | |
| Alkhudaydi et al. | An exploration of deep-learning based phenotypic analysis to detect spike regions in field conditions for UK bread wheat | |
| CN113780296A (en) | Remote sensing image semantic segmentation method and system based on multi-scale information fusion | |
| US20220215656A1 (en) | Method, apparatus, device for image processing, and storage medium | |
| Hamada et al. | Multi-spectral image segmentation based on the K-means clustering | |
| Xie et al. | Toward robotic weed control: detection of nutsedge weed in bermudagrass turf using inaccurate and insufficient training data | |
| Cardoso et al. | Classifying the content of social media images to support cultural ecosystem service assessments using deep learning models | |
| CN113435254A (en) | Sentinel second image-based farmland deep learning extraction method | |
| Wang et al. | CSA-CDGAN: channel self-attention-based generative adversarial network for change detection of remote sensing images | |
| CN112270671B (en) | Image detection method, device, electronic equipment and storage medium | |
| CN114092804A (en) | Method and device for identifying remote sensing image | |
| Oehmcke et al. | Deep point cloud regression for above-ground forest biomass estimation from airborne LiDAR | |
| CN114648706B (en) | Forest tree species identification method, device and equipment based on satellite remote sensing image | |
| CN116740422A (en) | Remote sensing image classification method and device based on multi-mode attention fusion technology | |
| CN111738972A (en) | Building detection system, method and device | |
| Nguyen et al. | Land cover classification at the wildland urban interface using high-resolution satellite imagery and deep learning | |
| CN113411550B (en) | Video coloring method, device, equipment and storage medium | |
| Chen et al. | CropQuant-Air: an AI-powered system to enable phenotypic analysis of yield-and performance-related traits using wheat canopy imagery collected by low-cost drones | |
| CN114462559A (en) | Target positioning model training method, target positioning method and device | |
| Huang et al. | Extraction of pine wilt disease based on a two-stage unmanned aerial vehicle deep learning method | |
| CN111310511A (en) | Method and device for identifying objects | |
| Sawant et al. | Sen-2 LULC: Land use land cover dataset for deep learning approaches | |
| Adesuyi | Automating land cover classification using time series NDVI: a case study in the berg river catchment area |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |