CN117422711B - Ocean vortex hyperspectral change detection method, device, equipment and medium - Google Patents
Ocean vortex hyperspectral change detection method, device, equipment and medium Download PDFInfo
- Publication number
- CN117422711B CN117422711B CN202311716239.4A CN202311716239A CN117422711B CN 117422711 B CN117422711 B CN 117422711B CN 202311716239 A CN202311716239 A CN 202311716239A CN 117422711 B CN117422711 B CN 117422711B
- Authority
- CN
- China
- Prior art keywords
- hyperspectral
- vortex
- ocean
- information
- spectrum
- 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
- 230000008859 change Effects 0.000 title claims abstract description 87
- 238000001514 detection method Methods 0.000 title claims abstract description 87
- 238000001228 spectrum Methods 0.000 claims abstract description 86
- 238000010586 diagram Methods 0.000 claims abstract description 50
- 238000000034 method Methods 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 22
- 230000007246 mechanism Effects 0.000 claims abstract description 14
- 230000003595 spectral effect Effects 0.000 claims description 29
- 238000000605 extraction Methods 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 24
- 230000004913 activation Effects 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 7
- 230000009977 dual effect Effects 0.000 claims description 5
- 230000002123 temporal effect Effects 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000006835 compression Effects 0.000 claims description 2
- 238000007906 compression Methods 0.000 claims description 2
- 239000000126 substance Substances 0.000 abstract description 6
- 238000004364 calculation method Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 7
- 238000010606 normalization Methods 0.000 description 4
- 238000000701 chemical imaging Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 206010039203 Road traffic accident Diseases 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000003260 vortexing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- 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
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Remote Sensing (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a method, a device, equipment and a medium for detecting ocean vortex hyperspectral change, wherein the method acquires a double-time-sequence vortex hyperspectral image to be detected; inputting the double-time-sequence vortex hyperspectral image to be detected into a pre-trained ocean vortex hyperspectral change detection model, and outputting an ocean vortex change gray level diagram; the ocean vortex hyperspectral change detection model is obtained after training based on double-time-sequence ocean vortex hyperspectral image sample data and a predetermined gray level map label. According to the ocean vortex hyperspectral change detection method provided by the invention, the attention to the spatial difference of similar spectrum substances and different spectrum substances is increased by using the spatial significance information attention mechanism based on the center pixel distance and the spectrum similarity, the attention to ocean vortex is enhanced, the attention to the background ocean is reduced, and the accuracy of ocean vortex hyperspectral change detection is further improved.
Description
Technical Field
The invention relates to the technical field of intelligent ocean and computer vision, in particular to a method, a device, equipment and a medium for detecting ocean vortex hyperspectral change.
Background
The hyperspectral change detection of the ocean vortex is an important technology for timely identifying the ocean vortex change by utilizing a double-phase hyperspectral image. The main purpose of the method is to know the ocean vortex, research the ocean vortex and protect the ocean traffic and navigation safety, so the method has very important strategic significance. The hyperspectral image is different from many other remote sensing images, has hundreds of continuous spectral bands from ultraviolet to infrared, can capture spectral reflection or absorption information in an underwater environment through hyperspectral imaging, has very high spectral and spatial resolution, is favorable for excavating the position and structural characteristics of ocean vortex in space, and is widely applied to the fields of ocean science such as ocean environment monitoring, underwater topography and ecosystem research, underwater resource exploration, ocean disaster monitoring and the like.
Conventional methods for detecting vortices using remote sensing data are typically based on physical parameters, geometric features, and manual features. Although the data of the change of the ocean vortex at different times can be continuously provided, the cost is too high, the time is too long, and the information cannot be provided in time. The method for detecting the ocean vortex change from the hyperspectral image of the sea surface based on the deep learning technology is focused on. For the marine vortex hyperspectral change detection task, lack of labeled training data has been a typical problem, and for the extraction of hyperspectral image space context information, a Patch-based method is generally adopted, and the method focuses more on local information in limited space information. However, the limited acceptance field results in some loss of global information, and the expansion of the acceptance field results in excessive neighborhood information interfering with the judgment of center pixel detection. For the spectral characteristics of hyperspectral images, which have a broad band, each band contributes differently to the detection of changes. However, the band that does not contribute much to the change detection affects accuracy, and excessive data can stress the computer calculation, which can be a problem with redundant spectral information. For ocean vortexing, it is difficult to detect using only spectral/spatial/temporal features, and correlation between these information needs to be considered.
Therefore, a new method for detecting the hyperspectral change of ocean vortex is needed to solve the above problems.
Disclosure of Invention
In view of the foregoing, it is desirable to provide a method, apparatus, device and medium for detecting the hyperspectral variation of ocean vortex, which are used for solving the problem of inaccurate detection of the hyperspectral variation of ocean vortex in the prior art.
In order to solve the above problems, in a first aspect, the present invention provides a method for detecting hyperspectral variation of ocean vortex, comprising:
acquiring a double-time-sequence vortex hyperspectral image to be detected;
inputting the double-time-sequence vortex hyperspectral image to be detected into a pre-trained ocean vortex hyperspectral change detection model, and outputting an ocean vortex change gray level diagram; the ocean vortex hyperspectral change detection model is obtained after training based on double-time-sequence ocean vortex hyperspectral image sample data and a predetermined gray level map label.
Preferably, inputting the to-be-detected dual-timing-sequence vortex hyperspectral image into a pre-trained ocean vortex hyperspectral change detection model, and further including:
acquiring an ocean vortex hyperspectral image data set and preprocessing to obtain double-time-sequence ocean vortex hyperspectral image sample data;
constructing an ocean vortex hyperspectral change detection model based on the double-time-sequence ocean vortex hyperspectral image sample data;
and training the ocean vortex hyperspectral change detection model to obtain a trained ocean vortex hyperspectral change detection model.
Preferably, the ocean vortex hyperspectral change detection model comprises a space significance information enhancement module, a weight sharing space feature extraction module, a compact advanced spectrum information Tokenizer, a weight non-sharing time sequence-spectrum-space information joint extraction module and a prediction module; wherein,
the spatial saliency information enhancement module is used for processing initial spatial information of the double-time-sequence vortex hyperspectral image to be detected based on spatial attention, and extracting spatial saliency information;
the weight sharing spatial feature extraction module is used for extracting spatial features based on the spatial significance information to obtain a double-time sequence spatial feature map;
the compact advanced spectrum information token is used for performing spectrum characteristic information conversion on the double-time sequence space characteristic diagram to obtain an advanced spectrum token set containing redundancy-removing neighborhood compact information; wherein token represents the minimum unit in the text;
the weight unshared time sequence-spectrum-space information joint extraction module is used for extracting joint characteristics of time sequence-spectrum-space information from the advanced spectrum token set to obtain a double-time sequence token with joint significance information;
and the prediction module is used for carrying out pixel level prediction on the double-time sequence token with the joint significance information to obtain an ocean vortex change gray level diagram.
Preferably, the spatial saliency information enhancement module is specifically configured to:
the method comprises the steps of (1) enabling a double-time-sequence vortex hyperspectral image to be detected to pass through a maximum pooling layer and an average pooling layer aiming at the dimension of a spectrum channel, and searching for differences among spectrums by splicing and carrying out point convolution compression on spectrum information by 1X 1 to obtain a first characteristic image and a second characteristic image;
processing the first feature map and the second feature map based on a central spectrum pixel focusing mechanism to obtain a first feature map, wherein the formula is as follows:
wherein,representing the first profile, +_>Representing the second profile, +_>Representing said first attention profile, ++>Representing a fully connected layer; />Representing uniform sampling; />Representing a sigmoid activation function;
processing the first feature map and the second feature map based on a spectrum-like pixel focusing mechanism to obtain a first feature map, wherein the formula is as follows:
；
wherein,representing a second attention seeking, ->Representing a mask map->Center pixel point for representing feature mapbIs a spectral feature value of (1);
and respectively giving weights to the first attention force diagram and the second attention force diagram, then carrying out matrix addition to obtain a target attention force diagram, and multiplying the target attention force diagram by the double-time-sequence vortex hyperspectral image to be detected to obtain a double-time-sequence space feature diagram with significance.
Preferably, the compact advanced spectrum information Tokenizer is specifically used for:
separating the dual temporal spatial feature map from the center pixel and background neighboring pixels in a compact advanced spectral information Tokenizer usingThe relation between the spectrums of each pixel point in the background is learned through two full-connection layers, and the advanced spectrum characteristics of the center pixel are obtained;
for the advanced spectral features, deriving a spectral advanced feature channel attention map using stitching and a Softmax activation function in the channel dimension;
multiplying the spectral high-level feature channel attention map with the dual-time sequence space feature map, so that each point in the token contains neighborhood high-level feature spectrum information, and obtaining a first high-level spectrum token set and a second high-level spectrum token set which contain redundancy-removing neighborhood compact information.
Preferably, the weight non-shared time sequence-spectrum-space information joint extraction module is specifically configured to:
respectively extracting time sequence information and spectrum sequence information through two preset encoders, and correlating time sequence-spectrum information through the first high-level spectrum token set and the second high-level spectrum token set to obtain a token set with time sequence information and spectrum information;
combining the double-time sequence space feature diagram obtained in the convolution network with the token set with time sequence information and spectrum information through Cross-Attention based on four preset decoders to obtain a CLS token with double-time sequence combined significance information; CLS token represents a classification feature.
Preferably, the training of the ocean vortex hyperspectral variation detection model specifically includes:
and after the sample data of the double-time-sequence ocean vortex hyperspectral image are input into the ocean vortex hyperspectral change detection model, a loss function value is calculated, counter propagation is carried out, and weights of the full-connection layer and the convolution layer are optimized through a selected optimizer and corresponding parameters until a preset effect is achieved.
In a second aspect, the present invention also provides an ocean vortex hyperspectral change detection device, including:
the acquisition module is used for acquiring a double-time-sequence vortex hyperspectral image to be detected;
the detection module is used for inputting the double-time-sequence vortex hyperspectral image to be detected into a pre-trained ocean vortex hyperspectral change detection model and outputting an ocean vortex change gray level diagram; the ocean vortex hyperspectral change detection model is obtained after training based on double-time-sequence ocean vortex hyperspectral image sample data and a predetermined gray level map label.
In a third aspect, the invention also provides an electronic device comprising a memory and a processor, wherein,
the memory is used for storing programs;
the processor is coupled to the memory and is configured to execute the program stored in the memory, so as to implement the steps in the method for detecting the hyperspectral variation of the ocean vortex in any implementation manner.
In a fourth aspect, the present invention also provides a computer readable storage medium storing a computer readable program or instructions which, when executed by a processor, enable the implementation of the steps of a method for detecting hyperspectral variation of ocean vortex in any one of the above-mentioned implementations.
The beneficial effects of adopting the embodiment are as follows: according to the ocean vortex hyperspectral change detection method provided by the invention, the attention to the spatial difference of similar spectrum substances and different spectrum substances is increased by using the spatial significance information attention mechanism based on the center pixel distance and the spectrum similarity, the attention to ocean vortex is enhanced, the attention to the background ocean is reduced, and the accuracy of ocean vortex hyperspectral change detection is further improved.
Drawings
FIG. 1 is a flow chart of a method for detecting the hyperspectral changes of ocean vortex, which is provided by the invention;
FIG. 2 is a diagram of an ocean vortex hyperspectral variation detection model established by the invention;
FIG. 3 is a block diagram of a spatial saliency information enhancement module provided by the present invention;
FIG. 4 is a flow chart of a compact high-level spectral information Tokenizer process provided by the present invention;
FIG. 5 is a block diagram of a weight-unshared time-spectrum-space information joint extraction module according to the present invention;
FIG. 6 is a block diagram illustrating an embodiment of an ocean vortex hyperspectral variation detection device according to the present invention;
fig. 7 is a block diagram of an embodiment of an electronic device according to the present invention.
Detailed Description
Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form a part hereof, and together with the description serve to explain the principles of the invention, and are not intended to limit the scope of the invention.
Fig. 1 is a flowchart of a method for detecting the hyperspectral change of ocean vortex. As shown in fig. 1, the ocean vortex hyperspectral change detection method includes:
step 110, obtaining a double-time-sequence vortex hyperspectral image to be detected;
step 120, inputting the double-time-sequence vortex hyperspectral image to be detected into a pre-trained ocean vortex hyperspectral change detection model, and outputting an ocean vortex change gray level diagram; the ocean vortex hyperspectral change detection model is obtained after training based on double-time-sequence ocean vortex hyperspectral image sample data and a predetermined gray level map label.
Specifically, in step 110, an ocean vortex hyperspectral image is acquired through a hyperspectral imaging device, ocean vortex hyperspectral images of different times are photographed, and a double-time-sequence ocean vortex hyperspectral image to be detected is obtained.
Before executing step 120, the marine vortex hyperspectral image sample data and the predetermined gray scale map label are utilized to construct a marine vortex hyperspectral variation detection model, and training of the model is completed. Then, in step 120, the image to be predicted is input into a trained ocean vortex hyperspectral variation detection model, so that an ocean vortex variation gray scale image can be obtained, and ocean vortex hyperspectral variation detection is completed.
According to the ocean vortex hyperspectral change detection method provided by the invention, the attention to the spatial difference of similar spectrum substances and different spectrum substances is increased by using the spatial significance information attention mechanism based on the center pixel distance and the spectrum similarity, the attention to ocean vortex is enhanced, the attention to the background ocean is reduced, and the accuracy of ocean vortex hyperspectral change detection is further improved.
Based on the above embodiment, before the to-be-detected dual-timing-sequence vortex hyperspectral image is input into the pre-trained ocean vortex hyperspectral change detection model in step 120, the ocean vortex hyperspectral change detection method provided by the embodiment of the present invention further includes:
step 101, acquiring an ocean vortex hyperspectral image data set and preprocessing to obtain double-time-sequence ocean vortex hyperspectral image sample data;
and 102, constructing an ocean vortex hyperspectral change detection model based on the double-time-sequence ocean vortex hyperspectral image sample data.
And step 103, training the ocean vortex hyperspectral change detection model to obtain a trained ocean vortex hyperspectral change detection model.
Specifically, before the ocean vortex hyperspectral change detection, an ocean vortex hyperspectral change detection model needs to be constructed and trained. In order to construct an ocean vortex hyperspectral change detection model, firstly, an ocean vortex hyperspectral image dataset is collected and preprocessed to obtain double-sequence ocean vortex hyperspectral image sample data, in step 101, ocean vortex hyperspectral images are acquired through hyperspectral imaging equipment, ocean vortex hyperspectral images at different times are shot, gray level images of ocean vortex natural scene changes are acquired through manual annotation, gray level image labels corresponding to the ocean vortex hyperspectral images are added to obtain the double-sequence ocean vortex hyperspectral image sample data, and the double-sequence ocean vortex hyperspectral image sample data comprises the double-sequence ocean vortex hyperspectral images and the gray level image labels corresponding to the double-sequence ocean vortex hyperspectral images. Selecting a patch-based method for the double-time-sequence hyperspectral image, taking neighborhood pixels for each central pixel to form a sample for training, wherein C and p are the number of spectral bands and the size of the sample respectively, and finally forming the processed patch into a double-time-sequence training set.
Further, an ocean vortex hyperspectral change detection model is constructed based on the double-time-sequence ocean vortex hyperspectral image sample data, and finally, the ocean vortex hyperspectral change detection model is trained to obtain a trained ocean vortex hyperspectral change detection model.
Fig. 2 is a framework diagram of an ocean vortex hyperspectral variation detection model established by the invention, and as shown in fig. 2, the model comprises a spatial significance information enhancement module, a weight sharing spatial feature extraction module, a compact high-level spectrum information token, a weight non-sharing time sequence-spectrum-spatial information joint extraction module and a prediction module, wherein,
the spatial saliency information enhancement module is used for processing initial spatial information of the double-time-sequence vortex hyperspectral image to be detected based on spatial attention, and extracting spatial saliency information;
the weight sharing spatial feature extraction module is used for extracting spatial features based on the spatial significance information to obtain a double-time sequence spatial feature map;
the compact advanced spectrum information token is used for performing spectrum characteristic information conversion on the double-time sequence space characteristic diagram to obtain an advanced spectrum token set containing redundancy-removing neighborhood compact information; wherein, token is a converter that converts spatial features into salient spectral features; token represents the smallest unit in text.
The weight unshared time sequence-spectrum-space information joint extraction module is used for extracting joint characteristics of time sequence-spectrum-space information from the advanced spectrum token set to obtain a double-time sequence token with joint significance information;
and the prediction module is used for carrying out pixel level prediction on the double-time sequence token with the joint significance information to obtain an ocean vortex change gray level diagram.
On the basis of the above embodiment, the spatial saliency information enhancement module processes initial spatial information of the dual-temporal vortex hyperspectral image to be detected based on spatial attention, and extracts spatial saliency information, and specifically includes:
the double-time-sequence vortex hyperspectral image to be detected passes through a maximum pooling layer and an average pooling layer aiming at the dimension of a spectrum channel, spectrum information is compressed through splicing and convolution of a 1 multiplied by 1 point, and spectrum difference is searched for, so that the spectrum difference is obtainedA feature map and a second feature map;
processing the first feature map and the second feature map based on a central spectrum pixel attention mechanism to obtain a first attention map, wherein the formula is as follows:
wherein,representing the first profile, +_>Representing said second profile, wherein ∈>，；/>Two ocean vortex hyperspectral images representing double time sequences are respectively input into two feature images obtained by a convolution network. />Representing said first attention profile, ++>Representing a fully connected layer; />Representing uniform sampling; />Representing a sigmoid activation function;
processing the first feature map and the second feature map based on a spectrum-like pixel attention mechanism to obtain a second attention map, wherein the formula is as follows:
；
wherein,representing a second attention profile;
matrix addition is performed by giving weight to obtain a final attention diagram, and the final attention diagram is multiplied by a base sample to obtain a double-time sequence space characteristic diagram with significance.
Fig. 3 is a frame diagram of a spatial saliency information enhancement module provided by the present invention, and as shown in fig. 3, the spatial saliency information enhancement module of the ocean vortex hyperspectral variation detection model is specifically configured to:
using spatial attention, a dual temporal vortex hyperspectral image Patch to be detectedAnd Patch->Processing the initial spatial information of (a) wherein ± is>）/>P represents the height and width of Patch and C represents the number of channels. Then, key space features of the salient center and spectrum-like pixel points are reserved, space significance information is extracted, and images are restrainedAnd the characteristic representation of the information irrelevant to the change detection in the spatial characteristic.
The initial double-time sequence hyperspectral Patch passes through a maximum pooling layer and an average pooling layer aiming at the dimension of a spectrum channel, spectrum information is compressed through splicing and convolution through a 1X 1 point, spectrum differences are searched, and a first characteristic diagram is obtainedAnd a second characteristic map->The calculation formula is as follows:
wherein,a convolution layer representing a kernel size of 3 x 3, filled with 1;
in the center spectral pixel focus mechanism, a uniform sampling function is introduced in order to gradually reduce the focus level from the center to the periphery. For the obtained first characteristic diagramAnd a second characteristic map->To satisfy the input form, it is first necessary to makeRemodelling to a one-dimensional vector->And->And through the full connection layer, wherein->，/>. Furthermore, given the non-positive and monotonically decreasing probability of uniform sampling slope leading to negative numbers, the activation is performed using the Softplus activation function, the calculation formula is as follows:
wherein,and->Weight bias representing the full connection layer; />Representing Softplus activation functions.
For a uniform sampling function, to prevent the gradient from disappearing, the distance to the square root needs to be measured, and the calculation formula is as follows:
wherein (x, y) is the spatial coordinates of the pixel relative to the center pixel; k is the rate of descent; b is the value of the center pixel. At this time, k controls the attention rate, and the full link layer can learn the dropping rate k corresponding to each Patch, thereby realizing adaptive attention.
By passing throughSampling may result in contributions of each of the position pixels to the change detection. For the resulting vector->And->Remodelling to->. In actual calculation, note that the scoring range should be 0-1, so that the sigmoid activation function is used for activation, and the calculation formula is as follows:
wherein the method comprises the steps ofRepresenting a first attention attempt->Representing a fully connected layer; />Representing uniform sampling; />Representing a sigmoid activation function.
In the spectrum-like pixel focusing mechanism, in order to ensure that the focusing scoring range is within 0-1, a sigmoid function needs to be used first. The obtained characteristic diagram is then used forAnd->Center pixel of +.>Is extended toAnd (2) feature map->And comparing, and calculating the absolute value of the obtained feature map. Finally, the absolute value of the feature map is compared with a mask pattern filled with 1MASKThe subtraction, resulting in a second attention-seeking diagram,MASK/>. The higher the attention score, the greater the spectral similarity between the two, calculated as follows:
obtaining a first attention attempt after passing through two attention mechanismsAnd a second attention attemptAnd respectively giving weights to the first attention force diagram and the second attention force diagram, then carrying out matrix addition to obtain a target attention force diagram, multiplying the target attention force diagram by a double-time-sequence vortex hyperspectral image to be detected to obtain a double-time-sequence space feature diagram with significance, wherein the calculation formula is as follows:
wherein,；/>；/>representing a target attention attempt->Representing a central spectral pixel focus mechanism; />Representing a spectrum-like pixel focus mechanism; />Representing a matrix sum; />Representing a matrix element-by-element multiplication; />Representing the different weights of the two methods.
The spatial feature extraction module for weight sharing of the ocean vortex hyperspectral change detection model obtains a double-time sequence spatial feature map after the spatial significance information extraction module passes through the spatial significance information extraction moduleAnd->Respectively, to a convolution layer portion composed of three consecutive convolution blocks, the above-mentioned convolution block formula can be expressed as:
wherein,a point convolution operation representing a filter size of 1 x 1; />Representing a depth convolution operation with a filter size of 3 x 3; BN represents the normalization layer; reLU stands for activate function.
The invention extracts the significance information related to the change detection through CNN for the significance space feature extraction and a transducer architecture for frequency spectrum redundancy removal. And the accuracy of ocean vortex hyperspectral change detection can be further improved through the combination of a transducer encoder and a decoder and time-frequency spectrum-space significance characteristic information.
FIG. 4 is a flowchart of a compact high-level spectrum information Tokenizer provided by the present invention, as shown in FIG. 4, where the high-level spectrum information Tokenizer provided by the present invention is specifically used for:
dual temporal spatial feature map through weight sharing spatial feature extraction moduleAndafter that, the dual timing space feature map +.>And->Input to a token, a set of high-level spectra token containing redundant neighborhood-specific information is obtained as a preparatory transformation to input to a converter. Wherein the set of higher-order spectrum Token comprises a first higher-order spectrum Token set->And a second higher order spectrum Token set->The method comprises the steps of carrying out a first treatment on the surface of the p represents the width and height of the profile, c represents the number of channels of the profile, (-)>，/>）/>
Specifically, first, a dual-timing spatial signature is mapped in a compact high-level spectral information TokenizerAnd->Separate from the central pixel and the background neighboring pixels, use +.>A background pixel spectrum, wherein the relation between the spectrums of each pixel point in the background is learned through two fully connected layers, and the background pixel spectrum is fully connected with the background pixel spectrumIn the connecting layer, the channel number is reduced from c to L, and similarly, for the center pixel, the advanced spectral characteristics of the center pixel are obtained.
Then, for the resulting advanced spectral features, a spectral advanced feature channel attention map is obtained using stitching and a Softmax activation function in the channel dimension，/>。
Further, multiplying the spectrum advanced feature channel attention map with the dual-time sequence space feature map to make each point in the Token contain neighborhood advanced feature spectrum information to obtain a first advanced spectrum Token set containing redundancy-removing neighborhood compact informationAnd a second higher order spectrum Token set->. The calculation formula is as follows:
in the method, in the process of the invention,representing two fully connected layers; />Representing a sigmoid activation function; />Representing a matrix multiplication.
The invention utilizes the conversion process of the compact spectrum saliency information advanced spectrum information token to help to extract the advanced spectrum conceptual features, and removes the spectrum redundancy by combining the neighborhood information, thereby focusing on the effective wave band for the change detection.
Fig. 5 is a frame diagram of a weight non-shared time-spectrum-space information joint extraction module provided by the present invention, and as shown in fig. 5, the weight non-shared time-spectrum-space information joint extraction module provided by the present invention specifically includes:
two encoders and four decoders, wherein the two encoders extract time-series information and spectral-series information, respectively, through Token setAnd Token set->Correlating the time-spectral information to obtain Token set with time-spectral information, comprising +.>And->. Wherein (-) is>，/>）/>Token set represents a collection of tokens, and Token represents the smallest unit in text. Double-timing spatial feature map obtained in convolutional network by Cross-Attention in four decoders +.>And->And ∈k obtained from encoder>、/>The combination of time sequence and spectrum to obtain the double-time sequence with the combined displayCLS token of authoring information, including +.>And->. Wherein (-) is>，/>）/>CLS token represents a classification feature.
Specifically, a first high-order spectrum Token set is obtainedAnd a second higher order spectrum Token set->After that, token set is first added>And Token set->Splicing to obtain->It is input into the Transformers encoder architecture. The transducer encoder is mainly composed of three main modules: position coding, multi-headed self-attention layers and multi-layered perceptrons. After position coding, the encoder is used to model the context of the time series, the output shape of the first encoder is unchanged.
After remolding, obtainAgain position coding, modeling the context of the optical sequence using an encoder, and finally obtaining by the encoderA set of dual-timing spectrum Token set->And->。
In the encoder, the normalization layer occurs before the MSA/MLP, andor->Input to an encoder consisting of an alternation of MSA and MLP. At each layer, the calculation formula is as follows:
where LN represents the normalization layer.
The MSA input is input byOr->A triplet (query Q, key K, value V) is calculated, where the formula for MSA can be expressed as:
in the method, in the process of the invention,，/>,/>is a learnable parameter of three linear projection layers; d represents the channel dimension of the triplet.
The difference of the transducer decoder is that the multi-head self-attention layer is replaced by multi-head cross-attention layer, and the double-time sequence space characteristic diagram obtained by the convolution network is obtained,/>And token set with timing information and spectral information->，/>And respectively combining to obtain the joint characteristics. First, the spatial signature is reshaped in the decoder to obtain +.>，/>Then add CLS token for pixel level classification, perform position coding, input into decoder, output shape size and input +.>,/>Similarly, the added CLS token is taken out to obtain the CLS token with joint significance information>And CLS token->。
In the decoder, the normalization layer precedes each MA/MLP. Q in triplets in cross-attention mechanism comes from spatial features，/>K and V are from->And->The formula for the differences can be expressed as:
the prediction module of the ocean vortex hyperspectral change detection model obtains the CLS token with the joint significance informationAnd CLS token->And carrying out pixel level prediction to obtain a gray level diagram of change detection. First to double time sequence tAnd (3) subtracting absolute values of the oken, and obtaining a final prediction probability result through the full connection layer and the Softmax activation function. The formula can be expressed as:
in the method, in the process of the invention,representing two fully connected layers; />Representing a sigmoid activation function.
Based on the above embodiment, in step 103, training the ocean vortex hyperspectral variation detection model specifically includes:
and after the sample data of the double-time-sequence ocean vortex hyperspectral image are input into the ocean vortex hyperspectral change detection model, a loss function value is calculated, counter propagation is carried out, and weights of the full-connection layer and the convolution layer are optimized through a selected optimizer and corresponding parameters until a preset effect is achieved.
The training process of the ocean vortex hyperspectral change detection model specifically comprises the following steps:
calculating a model loss function value L, and training a model through a binary cross entropy loss function, wherein the calculation formula of the model loss function L is as follows:
wherein n represents the number of samples;representing a real label, with a value of 0 or 1; />Representing the predicted probability value calculated by the Softmax function.
Fig. 6 is a block diagram of an embodiment of an ocean vortex hyperspectral variation detection device according to the present invention, and as shown in fig. 6, an embodiment of the present invention further provides an ocean vortex hyperspectral variation detection device 600, including:
the training model module 601 is used for pre-constructing an ocean vortex hyperspectral change detection model and completing training to obtain a trained ocean vortex hyperspectral change detection model, wherein the ocean vortex hyperspectral change detection model comprises a spatial significance information enhancement module, a weight sharing spatial feature extraction module, a compact advanced spectrum information Tokenizer, a weight non-sharing time sequence-spectrum-spatial information combined extraction module and a prediction module;
the detection module 602 is configured to input the dual-timing vortex hyperspectral image into the trained ocean vortex hyperspectral variation detection model, and output an ocean vortex variation gray scale map.
Fig. 7 is a block diagram of an embodiment of an electronic device according to the present invention, and as shown in fig. 7, the present invention further provides an electronic device 700, which may be a mobile terminal, a desktop computer, a notebook computer, a palm computer, a server, or other computing devices. The electronic device 700 comprises a processor 701 and a memory 702, wherein the memory 702 has stored thereon an ocean vortex hyperspectral change detection program 703.
The memory 702 may in some embodiments be an internal storage unit of a computer device, such as a hard disk or memory of a computer device. The memory 702 may also be an external storage device of a computer device in other embodiments, such as a plug-in hard disk provided on the computer device, a Smart Media Card (SMC), a secure digital (SecureDigital, SD) Card, a Flash memory Card (Flash Card), etc. Further, the memory 702 may also include both internal storage units and external storage devices of the computer device. The memory 702 is used for storing application software installed on the computer device and various types of data, such as program codes for installing the computer device. The memory 702 may also be used to temporarily store data that has been output or is to be output. In one embodiment, the marine vortex hyperspectral change detection program 703, when executed by the processor 701, performs the steps of:
acquiring a double-time-sequence vortex hyperspectral image to be detected;
inputting the double-time-sequence vortex hyperspectral image to be detected into a pre-trained ocean vortex hyperspectral change detection model, and outputting an ocean vortex change gray level diagram; the ocean vortex hyperspectral change detection model is obtained after training based on double-time-sequence ocean vortex hyperspectral image sample data and a predetermined gray level map label.
The processor 701 may be, in some embodiments, a central processing unit (Central Processing Unit, CPU), microprocessor or other data processing chip for executing program code or processing data stored in the memory 702, such as executing a bus accident cause identification program or the like.
The present embodiment also provides a computer-readable storage medium having stored thereon an ocean vortex hyperspectral variation detection program which, when executed by a processor, implements the steps of:
acquiring a double-time-sequence vortex hyperspectral image to be detected;
inputting the double-time-sequence vortex hyperspectral image to be detected into a pre-trained ocean vortex hyperspectral change detection model, and outputting an ocean vortex change gray level diagram; the ocean vortex hyperspectral change detection model is obtained after training based on double-time-sequence ocean vortex hyperspectral image sample data and a predetermined gray level map label.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the various embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention.
Claims (8)
1. The method for detecting the hyperspectral change of the ocean vortex is characterized by comprising the following steps of:
acquiring a double-time-sequence vortex hyperspectral image to be detected;
inputting the double-time-sequence vortex hyperspectral image to be detected into a pre-trained ocean vortex hyperspectral change detection model, and outputting an ocean vortex change gray level diagram; the ocean vortex hyperspectral change detection model is obtained after training based on double-time-sequence ocean vortex hyperspectral image sample data and a predetermined gray level map label;
the ocean vortex hyperspectral change detection model comprises a space significance information enhancement module, a weight sharing space feature extraction module, a compact advanced spectrum information Tokenizer, a weight non-sharing time sequence-spectrum-space information joint extraction module and a prediction module; wherein,
the spatial saliency information enhancement module is used for processing initial spatial information of the double-time-sequence vortex hyperspectral image to be detected based on spatial attention, and extracting spatial saliency information;
the weight sharing spatial feature extraction module is used for extracting spatial features based on the spatial significance information to obtain a double-time sequence spatial feature map;
the compact advanced spectrum information token is used for performing spectrum characteristic information conversion on the double-time sequence space characteristic diagram to obtain an advanced spectrum token set containing redundancy-removing neighborhood compact information; wherein token is a converter for converting spatial features into salient spectral features, token represents the smallest unit in text;
the weight unshared time sequence-spectrum-space information joint extraction module is used for extracting joint characteristics of time sequence-spectrum-space information from the advanced spectrum token set to obtain a double-time sequence token with joint significance information;
the prediction module is used for carrying out pixel level prediction on the double-time sequence token with the joint significance information to obtain an ocean vortex change gray level diagram;
the spatial saliency information enhancement module is specifically configured to:
the method comprises the steps of (1) enabling a double-time-sequence vortex hyperspectral image to be detected to pass through a maximum pooling layer and an average pooling layer aiming at the dimension of a spectrum channel, and searching for differences among spectrums by splicing and carrying out point convolution compression on spectrum information by 1X 1 to obtain a first characteristic image and a second characteristic image;
processing the first feature map and the second feature map based on a central spectrum pixel attention mechanism to obtain a first attention map, wherein the formula is as follows:
wherein,representing the first profile, +_>Representing the second profile, +_>Representing said first attention profile, ++>Representing a fully connected layer; />Representing uniform sampling; />Representing a sigmoid activation function;
based on a spectrum-like pixel attention mechanism, the first feature map and the second feature map are processed to obtain a second attention map, wherein the formula is as follows:
；
wherein,representing a second attention seeking, ->Representing a mask map->Center pixel point for representing feature mapbIs a spectral feature value of (1);
and respectively giving weights to the first attention force diagram and the second attention force diagram, then carrying out matrix addition to obtain a target attention force diagram, and multiplying the target attention force diagram by the double-time-sequence vortex hyperspectral image to be detected to obtain a double-time-sequence space feature diagram with significance.
2. The method for detecting the hyperspectral variation of ocean vortex according to claim 1, wherein the step of inputting the double-time-series vortex hyperspectral image to be detected into a pre-trained ocean vortex hyperspectral variation detection model, further comprises the steps of:
acquiring an ocean vortex hyperspectral image data set and preprocessing to obtain double-time-sequence ocean vortex hyperspectral image sample data;
constructing an ocean vortex hyperspectral change detection model based on the double-time-sequence ocean vortex hyperspectral image sample data;
and training the ocean vortex hyperspectral change detection model to obtain a trained ocean vortex hyperspectral change detection model.
3. The ocean vortex hyperspectral variation detection method according to claim 1 wherein the compact high-level spectral information Tokenizer is specifically used for:
separating the dual temporal spatial feature map from the center pixel and background neighboring pixels in a compact advanced spectral information Tokenizer usingThe relation between the spectrums of each pixel point in the background is learned through two full-connection layers, and the advanced spectrum characteristics of the center pixel are obtained;
for the advanced spectral features, deriving a spectral advanced feature channel attention map using stitching and a Softmax activation function in the channel dimension;
multiplying the spectral high-level feature channel attention map with the dual-time sequence space feature map, so that each point in the token contains neighborhood high-level feature spectrum information, and obtaining a first high-level spectrum token set and a second high-level spectrum token set which contain redundancy-removing neighborhood compact information.
4. The method for detecting the hyperspectral variation of ocean vortex according to claim 3, wherein the weight non-shared time sequence-spectrum-space information joint extraction module is specifically configured to:
respectively extracting time sequence information and spectrum sequence information through two preset encoders, and correlating time sequence-spectrum information through the first high-level spectrum token set and the second high-level spectrum token set to obtain a token set with time sequence information and spectrum information;
combining the double-time sequence space feature diagram obtained in the convolution network with the token set with time sequence information and spectrum information through Cross-Attention based on four preset decoders to obtain a CLS token with double-time sequence combined significance information; CLS token represents a classification feature.
5. The method for detecting the hyperspectral variation of ocean according to claim 2, wherein the training of the hyperspectral variation of ocean model specifically comprises:
and after the sample data of the double-time-sequence ocean vortex hyperspectral image are input into the ocean vortex hyperspectral change detection model, a loss function value is calculated, counter propagation is carried out, and weights of the full-connection layer and the convolution layer are optimized through a selected optimizer and corresponding parameters until a preset effect is achieved.
6. An ocean vortex hyperspectral change detection device, characterized by comprising:
the acquisition module is used for acquiring a double-time-sequence vortex hyperspectral image to be detected;
the detection module is used for inputting the double-time-sequence vortex hyperspectral image to be detected into a pre-trained ocean vortex hyperspectral change detection model and outputting an ocean vortex change gray level diagram;
the ocean vortex hyperspectral change detection model is obtained after training based on the double-time-sequence ocean vortex hyperspectral image sample data and a predetermined gray scale map label according to any one of claims 1 to 5.
7. An electronic device comprising a memory and a processor, wherein,
the memory is used for storing programs;
the processor, coupled to the memory, is configured to execute the program stored in the memory to implement the steps of a marine vortex hyperspectral variation detection method as claimed in any one of claims 1 to 5.
8. A computer readable storage medium storing a computer readable program or instructions which when executed by a processor is capable of carrying out the steps of a method of marine vortex hyperspectral change detection as claimed in any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311716239.4A CN117422711B (en) | 2023-12-14 | 2023-12-14 | Ocean vortex hyperspectral change detection method, device, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311716239.4A CN117422711B (en) | 2023-12-14 | 2023-12-14 | Ocean vortex hyperspectral change detection method, device, equipment and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117422711A CN117422711A (en) | 2024-01-19 |
| CN117422711B true CN117422711B (en) | 2024-03-26 |
Family
ID=89528638
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311716239.4A Active CN117422711B (en) | 2023-12-14 | 2023-12-14 | Ocean vortex hyperspectral change detection method, device, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117422711B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117671432B (en) * | 2024-01-31 | 2024-05-07 | 苏州元脑智能科技有限公司 | Method and device for training change analysis model, electronic equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113406057A (en) * | 2021-06-29 | 2021-09-17 | 江苏大学 | Method and system for rapidly detecting pesticide residues in food |
| CN113592786A (en) * | 2021-07-12 | 2021-11-02 | 浙江大学 | Deep learning-based detection method for mesoscale vortices in ocean |
| CN115131313A (en) * | 2022-06-27 | 2022-09-30 | 浙江大学 | Hyperspectral image change detection method and device based on Transformer |
| CN115187861A (en) * | 2022-07-13 | 2022-10-14 | 哈尔滨理工大学 | Hyperspectral image change detection method and system based on depth twin network |
| CN115512162A (en) * | 2022-10-08 | 2022-12-23 | 中国石油大学(华东) | Terrain classification method based on attention twin network and multi-mode fusion features |
| CN116342517A (en) * | 2023-03-20 | 2023-06-27 | 浙江大学 | Hyperspectral image change detection method and device based on spectrum-time transducer |
| CN117036879A (en) * | 2023-07-24 | 2023-11-10 | 湖南大学 | Method and device for identifying ground object by fusion of hyperspectral and laser radar multimode images and spatial spectrum |
| CN117115675A (en) * | 2023-08-31 | 2023-11-24 | 西安电子科技大学 | Cross-time-phase light-weight spatial spectrum feature fusion hyperspectral change detection method, system, equipment and medium |
-
2023
- 2023-12-14 CN CN202311716239.4A patent/CN117422711B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113406057A (en) * | 2021-06-29 | 2021-09-17 | 江苏大学 | Method and system for rapidly detecting pesticide residues in food |
| CN113592786A (en) * | 2021-07-12 | 2021-11-02 | 浙江大学 | Deep learning-based detection method for mesoscale vortices in ocean |
| CN115131313A (en) * | 2022-06-27 | 2022-09-30 | 浙江大学 | Hyperspectral image change detection method and device based on Transformer |
| CN115187861A (en) * | 2022-07-13 | 2022-10-14 | 哈尔滨理工大学 | Hyperspectral image change detection method and system based on depth twin network |
| CN115512162A (en) * | 2022-10-08 | 2022-12-23 | 中国石油大学(华东) | Terrain classification method based on attention twin network and multi-mode fusion features |
| CN116342517A (en) * | 2023-03-20 | 2023-06-27 | 浙江大学 | Hyperspectral image change detection method and device based on spectrum-time transducer |
| CN117036879A (en) * | 2023-07-24 | 2023-11-10 | 湖南大学 | Method and device for identifying ground object by fusion of hyperspectral and laser radar multimode images and spatial spectrum |
| CN117115675A (en) * | 2023-08-31 | 2023-11-24 | 西安电子科技大学 | Cross-time-phase light-weight spatial spectrum feature fusion hyperspectral change detection method, system, equipment and medium |
Non-Patent Citations (2)
| Title |
|---|
| matnet：a combining multi-attention and transformer network for hyperspectral image classification;Bo Zhang 等;《IEEE》;20230309;第61卷(第3期);第1-14页 * |
| 陈亚雄.基于深度学习的检索技术研究.《中国博士学位论文全文数据库信息科技辑》.第I138-81页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117422711A (en) | 2024-01-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jiang et al. | HDCB-Net: A neural network with the hybrid dilated convolution for pixel-level crack detection on concrete bridges | |
| CN110738207B (en) | Character detection method for fusing character area edge information in character image | |
| CN117422711B (en) | Ocean vortex hyperspectral change detection method, device, equipment and medium | |
| CN111079683A (en) | Remote sensing image cloud and snow detection method based on convolutional neural network | |
| Tian et al. | Multiscale building extraction with refined attention pyramid networks | |
| Li et al. | DMNet: A network architecture using dilated convolution and multiscale mechanisms for spatiotemporal fusion of remote sensing images | |
| CN110929080A (en) | Optical remote sensing image retrieval method based on attention and generation countermeasure network | |
| CN115375999B (en) | Target detection model, method and device applied to hazardous chemical vehicle detection | |
| CN114140831B (en) | Human body posture estimation method and device, electronic equipment and storage medium | |
| Chen et al. | MSF-Net: A multiscale supervised fusion network for building change detection in high-resolution remote sensing images | |
| CN115471675A (en) | Disguised object detection method based on frequency domain enhancement | |
| Shafique et al. | SSViT-HCD: A spatial–spectral convolutional vision transformer for hyperspectral change detection | |
| CN117475216A (en) | Hyperspectral and laser radar data fusion classification method based on AGLT network | |
| CN116524189A (en) | High-resolution remote sensing image semantic segmentation method based on coding and decoding indexing edge characterization | |
| Wang et al. | An unsupervised heterogeneous change detection method based on image translation network and post-processing algorithm | |
| Verma et al. | Survey on image copy-move forgery detection | |
| CN116206221B (en) | Water flare detection method and system | |
| CN111539435A (en) | Semantic segmentation model construction method, image segmentation equipment and storage medium | |
| Mo et al. | SAUNet3+ CD: A Siamese-attentive UNet3+ for change detection in remote sensing images | |
| Kekre et al. | Sectorization of Full Kekre’s Wavelet Transform for Feature extraction of Color Images | |
| CN116310474A (en) | End-to-end relationship identification method, model training method, device, equipment and medium | |
| CN115223181A (en) | Text detection-based method and device for recognizing characters of seal of report material | |
| CN112241765A (en) | Image classification model and method based on multi-scale convolution and attention mechanism | |
| CN117372789B (en) | Image classification method and image classification device | |
| CN110909914A (en) | Litigation success rate prediction method, litigation success rate prediction device, computer device, 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |