CN116310510A - Hyperspectral image classification method based on small sample deep learning - Google Patents

Hyperspectral image classification method based on small sample deep learning Download PDF

Info

Publication number
CN116310510A
CN116310510A CN202310095139.8A CN202310095139A CN116310510A CN 116310510 A CN116310510 A CN 116310510A CN 202310095139 A CN202310095139 A CN 202310095139A CN 116310510 A CN116310510 A CN 116310510A
Authority
CN
China
Prior art keywords
small sample
hyperspectral image
layer
feature extraction
neural network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310095139.8A
Other languages
Chinese (zh)
Inventor
高大化
孙科
张中强
刘丹华
牛毅
石光明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN202310095139.8A priority Critical patent/CN116310510A/en
Publication of CN116310510A publication Critical patent/CN116310510A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/58—Extraction of image or video features relating to hyperspectral data
    • 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
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • 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
    • Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
    • Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Remote Sensing (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a hyperspectral image classification method based on small sample deep learning, which comprises the following steps: acquiring a source domain hyperspectral image set and a target domain hyperspectral image set; extracting a plurality of first image blocks and a plurality of second image blocks by taking each pixel point of the source domain hyperspectral image and the target domain hyperspectral image after filling treatment as a center; in each category, randomly selecting a part of first image blocks to form a source domain support set and a part of first image blocks to form a source domain query set, and randomly selecting a part of second image blocks to form a target domain support set and a part of first image blocks to form a target domain query set; training the small sample spectral space feature extraction convolutional neural network by using a support set and a query set to obtain a trained network; and inputting the hyperspectral image to be classified into a trained small sample spectral space feature extraction convolutional neural network to obtain a classification result. The method and the device can improve the classification precision and extract the spatial information contained in the hyperspectral image in a deeper level.

Description

Hyperspectral image classification method based on small sample deep learning
Technical Field
The invention belongs to the technical field of remote sensing information processing, and relates to a hyperspectral image classification method based on deep learning of a small sample.
Background
The hyperspectral records the continuous spectrum characteristics of the ground object targets by the abundant wave band information, and has the possibility of identifying more ground object targets and classifying the targets with higher precision. Unlike common natural images, hyperspectral image data presents a three-dimensional structure, the spectrum information is quite rich, the space information is relatively less, and the key of the hyperspectral image classification technology is to classify sample types by utilizing the space characteristics and the inter-spectrum characteristics of hyperspectral images. However, there are few training samples of hyperspectral images, so that the problem of fitting is easy to occur for classification methods with a relatively large amount of required parameters, and how to train an efficient classification model is important for hyperspectral image classification under the condition of a small amount of samples.
Kun Tan et al in its published paper "a novel semi-supervised hyperspectral image classification approach based on spatial neighborhood information and classifier combination" (ISPRS journal of photogrammetry and remote sensing, 2015) propose a new semi-supervised HSI classification method that combines spatial neighborhood information with a classifier to enhance classification ability. The paper "Semi-supervised hyperspectral image classification via spatial-restricted self-training" published by Yue Wu et al (Remote Sensing, 2020) proposes a Semi-supervised method that uses self-training to progressively assign highly trusted pseudo-labels to unlabeled samples by clustering and uses spatial constraints to adjust the self-training process. However, these approaches assume that the marked and unmarked samples are from the same dataset, which means that the classification performance is still limited by the number of marked samples in the data to be classified (i.e. the target domain).
The paper "Deep few-shot learning for hyperspectral image classification" published by Bing Liu et al (IEEE Transactions on Geoscience and Remote Sensing, 2018) proposes a small sample Deep learning method to solve the small sample problem of HSI classification, which better aids classification by learning metric space from a training set. The Kuiliang Gao et al published paper "Deep relation network for hyperspectral image few-shot classification" (Remote Sensing, 2020) devised a new deep classification model based on relational networks and trained with the ideas of meta-learning.
Except the above-listed hyperspectral image classification method, the existing hyperspectral image classification method based on the deep convolutional neural network is similar to the above-mentioned method, and the commonality of the methods is that information is lost due to insufficient utilization rate of extracted features during extraction of inter-spectrum and spatial features, or information redundancy is caused by retaining too much irrelevant information, key information in hyperspectral image bands cannot be fully utilized, more resolvable spectral-empty semantic features are obtained, and a large number of hyperspectral samples are needed for training the neural network during training, so that the hyperspectral image classification effect of the methods is poor during training with few samples, and the difference of information between different spectrums is not focused deeply. And considering the problem of fewer hyperspectral datasets, although there are methods to solve the hyperspectral classification problem using small sample learning, no suitable method has been proposed to solve the domain adaptation problem caused by using different hyperspectral datasets. In classification, the loss function adopted by the method is too single, and the requirement of high-precision classification cannot be met.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a hyperspectral image classification method based on small sample deep learning, which helps classifying a target domain hyperspectral data set with a small number of label samples by learning priori knowledge in the source domain hyperspectral data set and using the source domain hyperspectral data set with a large number of label samples so as to improve the accuracy of classification of ground object targets in the hyperspectral image under the condition of training with a small number of samples. The technical problems to be solved by the invention are realized by the following technical scheme:
the embodiment of the invention provides a hyperspectral image classification method based on small sample deep learning, which comprises the following steps:
step 1, acquiring a source domain hyperspectral image set and a target domain hyperspectral image set, wherein the source domain hyperspectral image set comprises a plurality of Zhang Yuanyu hyperspectral images, and the target domain hyperspectral image set comprises a plurality of target domain hyperspectral images;
step 2, respectively filling the edge parts of the source domain hyperspectral image and the target domain hyperspectral image, and extracting a plurality of first image blocks and a plurality of second image blocks by taking each pixel point of the filled source domain hyperspectral image and target domain hyperspectral image as a center;
Step 3, in each category, randomly selecting part of the first image blocks to form a source domain support set and part of the first image blocks to form a source domain query set, and randomly selecting part of the second image blocks to form a target domain support set and part of the first image blocks to form a target domain query set;
step 4, based on a random gradient descent method, carrying out alternate training on a small sample spectral space feature extraction convolutional neural network by using the source domain support set, the source domain query set, the target domain support set and the target domain query set to obtain a trained small sample spectral space feature extraction convolutional neural network, wherein the small sample spectral space feature extraction convolutional neural network extracts features from two aspects of spectrum and space, and a total loss function of the small sample spectral space feature extraction convolutional neural network consists of a cross entropy loss function, a correlation alignment loss function and a maximum average difference loss function;
and step 5, inputting the hyperspectral image to be classified into the trained small sample spectral space feature extraction convolutional neural network to obtain a classification result.
In an embodiment of the invention, the step 2 includes:
step 2.1, filling pixels with pixel values of 0 into the peripheries of the source domain hyperspectral image and the target domain hyperspectral image respectively to obtain a filled source domain hyperspectral image and a filled target domain hyperspectral image;
And 2.2, selecting the first image block and the second image block with the space size of (2t+1) x (2t+1) and the channel number of d by taking each pixel point in the filled source domain hyperspectral image and the filled target domain hyperspectral image as a center, wherein t is an integer greater than 0.
In an embodiment of the invention, the structure of the small sample spectral space feature extraction convolutional neural network includes a spectral branch network, a spatial branch network, a domain attention module, a first splicing layer, a full connection layer and a softmax classifier, the spectral branch network, the spatial branch network are connected in parallel and then sequentially connected in series with the domain attention module, the first splicing layer, the full connection layer and the softmax classifier are connected in series, the spectral branch network includes 2 3D deformable convolution blocks and 2 first maximum pooling layers, the 1 st 3D deformable convolution block, the 1 st first maximum pooling layer, the 2 nd 3D deformable convolution block, the 2 nd first maximum pooling layer are sequentially connected in series, and the spatial branch network domain attention module includes a multi-scale spatial feature extraction module, a second splicing layer and a second maximum pooling layer which are sequentially connected in series.
In an embodiment of the present invention, the 3D deformable convolution block includes 3D deformable convolution layers and 3 first activation function layers, where the 1 st 3D deformable convolution layer, the 1 st first activation function layer, the 2 nd 3D deformable convolution layer, the 2 nd first activation function layer, the 3 rd 3D deformable convolution layer, and the 3 rd first activation function layer are sequentially connected in series, and an output of the 1 st first activation function layer and an output of the 3 rd 3D deformable convolution layer are added to form a residual structure.
In an embodiment of the invention, the multi-scale spatial feature extraction module includes 2 scale operation layers, 3 convolution layers, 3 normalization layers, and 3 second activation function layers;
the 1 st convolution layer, the 1 st normalization layer and the 1 st second activation function layer are sequentially connected in series;
the 1 st scale operation layer, the 2 nd convolution layer, the 2 nd normalization layer and the 2 nd second activation function layer are sequentially connected in series;
the 2 nd scale operation layer, the 3 rd convolution layer, the 3 rd normalization layer and the 3 rd second activation function layer are sequentially connected in series;
the 1 st second activation function layer, the 2 nd second activation function layer and the 3 rd second activation function layer are connected in parallel and then connected in series with the second splicing layer and the second maximum pooling layer in sequence.
In an embodiment of the invention, the total loss function of the small sample spectral space feature extraction convolutional neural network is:
L total ＝L fsl +L coral +L MMD
wherein ,Ltotal Representing the total loss function L of the small sample spectral space feature extraction convolutional neural network fsl Represents cross entropy loss, L coral Representing dependency alignment loss, L MMD Representing the maximum averaged delta loss;
cross entropy loss L fsl Expressed as:
Figure BDA0004085658520000051
Figure BDA0004085658520000052
wherein d (·) represents the Euclidean distance, F ω (. Cndot.) the feature extraction function of parameter ω, f l Features representing the first class in the source domain support set or the target domain support set, C being the number of classes, x j Representing one sample in a source domain query set or a target domain query set, y j Representing sample x j Q represents a source domain query set or a target domain query set;
correlation alignment loss L coral Expressed as:
Figure BDA0004085658520000061
wherein ,
Figure BDA0004085658520000062
frobenius norms, C representing matrix S Covariance matrix representing source domain features, C T A covariance matrix representing the characteristics of the target domain, and d represents the dimension of the characteristics;
maximum averaged difference loss L MMD Expressed as:
Figure BDA0004085658520000063
wherein ,
Figure BDA0004085658520000064
represents the spatial distance, phi (·) represents the mapping function, +.>
Figure BDA0004085658520000065
Representing source domain features, ++>
Figure BDA0004085658520000066
Representing the characteristics of the target domain, X s Representing a source domain dataset, X t Representing a target domain dataset, n s Representing the number of data in the source domain dataset, n t Representing the number of data in the target domain dataset.
In an embodiment of the invention, the step 4 includes:
setting the initial learning rate of training as alpha and the iteration times as T, and sending the source domain support set and the source domain query set into the small sample spectral space feature extraction convolutional neural network for training when the iteration is performed for odd times, and calculating the loss value between the features of the source domain support set and the features of the source domain query set by using the total loss function so as to update the parameters in the small sample spectral space feature extraction convolutional neural network; and when the number of iterations is even, the target domain support set and the target domain query set are sent into the small sample spectral space feature extraction convolutional neural network to train, the total loss function is utilized to calculate the loss value between the features of the target domain support set and the features of the target domain query set so as to update the parameters in the small sample spectral space feature extraction convolutional neural network until the loss value of the small sample spectral space feature extraction convolutional neural network is not reduced any more and the current training round number is smaller than the iteration number T or the training round number reaches the iteration number T, and training of the small sample spectral space feature extraction convolutional neural network is stopped to obtain the trained small sample spectral space feature extraction convolutional neural network.
In an embodiment of the invention, the small sample spectral space feature extracts a weight vector W updated by a convolutional neural network new The method comprises the following steps:
Figure BDA0004085658520000071
wherein ,Ltotal The method is characterized in that the method comprises the steps of representing a total loss function of a small sample spectral null feature extraction convolutional neural network, W represents a weight vector before updating the small sample spectral null feature extraction convolutional neural network, and R represents a learning rate.
Compared with the prior art, the invention has the beneficial effects that:
the hyperspectral image classification method aims at the abundant spectral characteristic information and spatial characteristic information of the hyperspectral image, and the hyperspectral image is classified by utilizing the small sample spectral space characteristic extraction convolutional neural network, wherein the small sample spectral space characteristic extraction convolutional neural network extracts characteristics from two aspects of spectrum and space respectively, so that the classification precision can be improved, and the spatial information contained in the hyperspectral image can be extracted more deeply.
According to the invention, in the spectrum branch, the 3D deformable convolution and residual error structure is adopted, so that the convolution kernel can better extract deep information for the hyperspectral image with an irregular shape, can reserve information of a shallow network, and improves classification precision; the spatial branch circuit adopts multi-scale operation, and for the input hyperspectral sample, two parts are copied first, and then edge pixel points are discarded in sequence, so that three input samples with different spatial resolutions are obtained, and the spatial information contained in the hyperspectral image can be extracted in a deeper layer.
Other aspects and features of the present invention will become apparent from the following detailed description, which refers to the accompanying drawings. It is to be understood that the drawings are designed solely for purposes of illustration and not as a definition of the limits of the invention, for which reference should be made to the appended claims. It should be further understood that the drawings are not necessarily drawn to scale and that, unless otherwise indicated, they are merely intended to conceptually illustrate the structures and procedures described herein.
Drawings
Fig. 1 is a schematic flow chart of a hyperspectral image classification method based on small sample deep learning according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a model structure of a convolutional neural network for small sample spectral space feature extraction according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a 3D deformable convolution module in a spectral branch according to an embodiment of the present disclosure;
FIG. 4 is a schematic structural diagram of a multi-scale spatial feature extraction module according to an embodiment of the present invention;
FIG. 5 is a simulation of classification results on a University of Pavia dataset with the present invention and existing two networks, respectively;
FIG. 6 is a simulation of the classification results on the Indian pins dataset with the present invention and the existing two networks, respectively.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but embodiments of the present invention are not limited thereto.
Example 1
The existing hyperspectral image classification methods based on the deep convolutional neural network have the defects that information is lost due to insufficient utilization rate of extracted features during inter-spectrum and space feature extraction, or information redundancy is caused by excessive irrelevant information, key information in hyperspectral image wave bands cannot be fully utilized, spectral-space semantic features with higher resolution are obtained, and a large number of hyperspectral samples are needed for training the neural network during training, so that the hyperspectral image classification effect of the methods is poor during training of few samples, and the difference of information among different spectrums is not focused deeply.
Based on this, the present invention proposes a hyperspectral image classification method based on small sample deep learning, referring specifically to fig. 1, fig. 1 is a schematic flow chart of a hyperspectral image classification method based on small sample deep learning provided by the embodiment of the present invention, and the hyperspectral image classification method based on small sample deep learning provided by the embodiment of the present invention specifically may include steps 1 to 4, where:
Step 1, acquiring a source domain hyperspectral image set and a target domain hyperspectral image set, wherein the source domain hyperspectral image set comprises a plurality of Zhang Yuanyu hyperspectral images, and the target domain hyperspectral image set comprises a plurality of target domain hyperspectral images.
Specifically, the hyperspectral image is a three-dimensional data S.epsilon.R h×w×c Each wave band in the hyperspectral image corresponds to a two-dimensional matrix S in the three-dimensional data i ∈R h×w Wherein e denotes a symbol belonging to the real number field, R denotes a real number field, h denotes a length of the hyperspectral image, w denotes a width of the hyperspectral image, c denotes a number of spectral bands of the hyperspectral image, i denotes a number of spectral bands in the hyperspectral image, i=1, 2, …,c。
In this embodiment, the source domain hyperspectral image set adopts a Chikusei dataset, and the target domain hyperspectral image set adopts a UP dataset.
And 2, respectively filling the edge parts of the source domain hyperspectral image and the target domain hyperspectral image, and extracting a plurality of first image blocks and a plurality of second image blocks by taking each pixel point of the filled source domain hyperspectral image and target domain hyperspectral image as a center.
Specifically, in this embodiment, the source domain hyperspectral image and the target domain hyperspectral image are respectively subjected to filling processing, so that a first image block and a second image block at the edge can also be obtained, and the first image block and the second image block at the edge contain required information.
And 2.1, filling pixels with pixel values of 0 into the peripheries of the source domain hyperspectral image and the target domain hyperspectral image respectively to obtain a filled source domain hyperspectral image and a filled target domain hyperspectral image.
And 2.2, selecting a first image block and a second image block with the space size of (2t+1) × (2t+1) and the channel number of d by taking each pixel point in the filled source domain hyperspectral image and the filled target domain hyperspectral image as a center, wherein t is an integer greater than 0, for example, the size of the image block is 9×9, and therefore t=4.
Specifically, the first image block is a pixel block extracted by taking a pixel point in the filled source domain hyperspectral image as a center, and the second image block is a pixel block extracted by taking a pixel point in the filled target domain hyperspectral image as a center. Wherein the number of channels d is the same as the number of spectral bands of the hyperspectral image.
And 3, randomly selecting part of the first image block composition source domain support set and part of the first image block composition source domain query set from each category, and randomly selecting part of the second image block composition target domain support set and part of the first image block composition target domain query set.
Specifically, the first image block and the second image block are distributed to the set to which the class belongs according to the class of the central pixel point, and the class is water, glass and the like. Selecting part of the source domain support sets from the first image blocks of each category, and selecting part of the source domain query sets; and selecting part of the second image blocks in each category to form a target domain support set, and selecting part of the second image blocks to form a target domain query set.
For example, a Chikusei data set is adopted in a source domain hyperspectral image set, 200 first image blocks are selected from each class to form a source domain data set, then an image block is selected from the source domain data set randomly for each class to form a source domain support set, and 19 first image blocks are selected randomly for each class to form a source domain query set; the target domain hyperspectral image set adopts UP and other data sets, each class selects 5 second image blocks to form a target domain data set, after data enhancement (namely copying of the target domain data set) is carried out, one image block is randomly selected from the target domain data set for each class to form a target domain support set, and each class randomly selects 19 image blocks to form a target domain query set.
And 4, based on a random gradient descent method, carrying out alternate training on the small sample spectral space feature extraction convolutional neural network by using a source domain support set, a source domain query set, a target domain support set and a target domain query set to obtain a trained small sample spectral space feature extraction convolutional neural network, wherein the small sample spectral space feature extraction convolutional neural network extracts features from two aspects of spectrum and space, and a total loss function of the small sample spectral space feature extraction convolutional neural network consists of a cross entropy loss function, a correlation alignment loss function and a maximum averaging difference loss function.
Specifically, referring to fig. 2, the structure of the small sample spectral space feature extraction convolutional neural network includes a spectral branch network, a spatial branch network, a domain attention module, a first splicing layer, a full connection layer and a softmax classifier, the spectral branch network and the spatial branch network are connected in parallel and then connected in series with the domain attention module, the first splicing layer, the full connection layer and the softmax classifier are sequentially connected in series, the spectral branch network includes 2 3D deformable convolution blocks and 2 first maximum pooling layers, the 1 st 3D deformable convolution block, the 1 st first maximum pooling layer, the 2 nd 3D deformable convolution block and the 2 nd first maximum pooling layer are sequentially connected in series, and the spatial branch network domain attention module includes a multi-scale spatial feature extraction module, a second splicing layer and a second maximum pooling layer which are sequentially connected in series.
The convolution kernel size of the 1 st first maximum pooling layer is set to 2×2×4, the number of convolution kernels is set to 8, the convolution kernel size of the second first maximum pooling layer is set to 2×2×4, and the number of convolution kernels is set to 16.
Referring to fig. 3, the 3D deformable convolution block includes 3D deformable convolution layers and 3 first activation function layers, where the 1 st 3D deformable convolution layer, the 1 st first activation function layer, the 2 nd 3D deformable convolution layer, the 2 nd first activation function layer, the 3 rd 3D deformable convolution layer, and the 3 rd first activation function layer are sequentially connected in series, and the output of the 1 st first activation function layer and the output of the 3 rd 3D deformable convolution layer are added to form a residual structure, and the 3D deformable convolution layer is a structure formed by adding a deformable convolution kernel in 3D convolution.
The convolution kernel sizes of the 3D deformable convolution layers are all set to be 3 x 3; the activation function of each first activation function layer is set as a ReLU activation function, expressed as follows:
ReLU(x)＝max(0,x)
where x represents the input of the activation function.
Referring to fig. 4, the multi-scale spatial feature extraction module includes 2 scale operation layers, 3 convolution layers, 3 normalization layers, and 3 second activation function layers, wherein:
the 1 st convolution layer, the 1 st normalization layer and the 1 st second activation function layer are sequentially connected in series;
the 1 st scale operation layer, the 2 nd convolution layer, the 2 nd normalization layer and the 2 nd second activation function layer are sequentially connected in series;
the 2 nd scale operation layer, the 3 rd convolution layer, the 3 rd normalization layer and the 3 rd second activation function layer are sequentially connected in series;
the 1 st second activation function layer, the 2 nd second activation function layer and the 3 rd second activation function layer are connected in parallel and then connected in series with the second splicing layer and the second maximum pooling layer in sequence.
The 1 st scale operation layer in the multi-scale space feature extraction module reduces one pixel point to the peripheral edge of the selected image block, the 2 nd scale operation layer reduces two pixel points to the peripheral edge of the selected image block, the convolution kernel size of the 1 st convolution layer is set to be 5 x 4, the convolution kernel size of the 2 nd convolution layer is set to be 3 x 4, the convolution kernel size of the 3 rd convolution layer is set to be 1 x 4, the number is set to be 16, and the activation function of each second activation function layer is set to be a ReLU activation function;
The output of the three second activation function layers is 16 features with the size of 5 x 25, 16 features with the size of 5 x 75 are obtained through the splicing operation of the second splicing layer, then the pooling operation is performed through the second maximum pooling layer, the convolution kernel of the second maximum pooling layer is set to be 2 x 8, and the number of the convolution kernels is set to be 16.
The domain attention module adopts 2D convolution layers and comprises 1 spectrum attention module and 2 space attention modules, wherein the spectrum attention module is 1 2D convolution layer, the 2 space attention modules are 2D convolution layers, the 3 2D convolution layers are sequentially connected in series, the size of convolution kernel of the spectrum attention module is 9*9, the number of convolution kernels is 1, and the domain attention module comprises space attention and inter-spectrum attention;
in the embodiment, after a spectrum branch network and a space branch network are connected in parallel, the spectrum branch network and the space branch network are connected in series with a domain attention module, a first splicing layer, a full connection layer and a softmax classifier to form a small sample spectrum space feature extraction convolution neural network, the domain attention module is a convolution layer, the domain attention module executes convolution operation to obtain a weighting coefficient, and the small sample spectrum space feature extraction convolution neural network selects a cross entropy loss function, a correlation alignment loss function and a maximum average difference loss function as loss functions of the small sample spectrum space feature extraction convolution neural network.
In this embodiment, the total loss function of the small sample spectral space feature extraction convolutional neural network is:
L total ＝L fsl +L coral +L MMD
wherein ,Ltotal Convolutional neural representing small sample spectral space feature extractionTotal loss function of network, L fsl Represents a cross entropy loss function, L coral Representing a correlation alignment loss function, L MMD Representing a maximum averaged difference loss function;
cross entropy loss function L fsl Expressed as:
Figure BDA0004085658520000131
Figure BDA0004085658520000132
wherein ,Lfsl Representing the loss value between the predicted and real label vectors, d (·) represents the Euclidean distance, F ω (. Cndot.) the feature extraction function of parameter ω, f l Features representing the first class in the source domain support set or the target domain support set, C being the number of classes, x j Representing one sample in a source domain query set or a target domain query set, y j Representing sample x j Q represents a source domain query set or a target domain query set.
Correlation alignment loss L coral Expressed as:
Figure BDA0004085658520000133
wherein ,
Figure BDA0004085658520000134
frobenius norms, C representing matrix S Covariance matrix representing source domain features, C T The covariance matrix representing the target domain features, wherein the source domain features are features spliced after the source domain support set and the source domain query set are extracted by a spectrum branch network and a space branch network, and the target domain features are features spliced after the target domain support set and the target domain query set are extracted by the spectrum branch network and the space branch network, and are spliced by the domain attention module and the first splicing layer Splicing the characteristics after the splicing of the layers, wherein d represents the dimension of the characteristics;
maximum averaged difference loss L MMD Expressed as:
Figure BDA0004085658520000141
wherein ,
Figure BDA0004085658520000142
represents the spatial distance measured by the mapping of data into the regenerated Hilbert space (RKHS), phi(s), which represents the mapping function, the->
Figure BDA0004085658520000143
Representing source domain features, ++>
Figure BDA0004085658520000144
Representing the characteristics of the target domain, X s Representing a source domain dataset, X t Representing a target domain dataset, n s Representing the number of data in the source domain dataset, n t Representing the number of data in the target domain dataset.
Based on the small sample spectral space feature extraction convolutional neural network and the total loss function thereof, the training method for the small sample spectral space feature extraction convolutional neural network comprises the following steps:
setting the initial learning rate of training as alpha and the iteration times as T, and sending a source domain support set and a source domain query set into a small sample spectral space feature extraction convolutional neural network for training when the iteration is performed for odd times, and calculating a loss value between the features of the source domain support set and the features of the source domain query set by using a total loss function so as to update parameters in the small sample spectral space feature extraction convolutional neural network; and when the number of iterations is even, the target domain support set and the target domain query set are sent into the small sample spectral space feature extraction convolutional neural network to train, the loss value between the features of the target domain support set and the features of the target domain query set is calculated by using the total loss function so as to update the parameters in the small sample spectral space feature extraction convolutional neural network until the loss value of the small sample spectral space feature extraction convolutional neural network is not reduced any more and the current training round number is smaller than the iteration number T or the training round number reaches the iteration number T, and training of the small sample spectral space feature extraction convolutional neural network is stopped to obtain the trained small sample spectral space feature extraction convolutional neural network.
Specifically, when iterating for odd times, a source domain support set and a source domain query set in a source domain data set are sent into a small sample spectral space feature extraction convolutional neural network for training, and loss values between features of the source domain support set and features of the source domain query set are calculated; and when iterating for even number, sending the target domain support set and the target domain query set in the target domain data set into the small sample spectral space feature extraction convolutional neural network for training, and equally calculating the loss value between the support set feature and the query set feature. And sequentially and alternately inputting a source domain data set and a target domain data set respectively to train the small sample spectral space feature extraction convolutional neural network, and continuously iterating to update parameters in the network.
The learning rate R of each input hyperspectral image block is set as follows: r=α.
Performing T times of weight update on the small sample spectral space feature extraction convolutional neural network to obtain an updated weight vector W new ：
Figure BDA0004085658520000151
wherein ,Ltotal The total loss function is represented, W represents the weight vector before updating of the small sample spectral null feature extraction convolutional neural network, and R represents the learning rate.
The next training sample set is input into a small sample spectral space feature extraction convolutional neural network, and the loss function value of the total loss function is updated to enable the loss function value L total Continuously decreasing until the loss function value L total The training of the network is stopped when the current training round frequency is smaller than the set iteration frequency T and the trained small sample spectral space feature extraction convolutional neural network is obtained; otherwise, when the number of training wheels reachesAnd T, stopping training the network to obtain the trained small sample spectral space feature extraction convolutional neural network.
And step 5, inputting the hyperspectral image to be classified into a trained small sample spectral space feature extraction convolutional neural network to obtain a classification result.
In this embodiment, in order to test the trained small sample spectral space feature extraction convolutional neural network, a test sample may be input to the trained small sample spectral space feature extraction convolutional neural network to obtain a class of the test sample, so as to complete classification of hyperspectral images.
Firstly, the spectrum branch network constructed by the invention can extract abundant inter-spectrum features through the 3D deformable convolution blocks, and the inter-spectrum features can be extracted with more resolution through focusing and screening the inter-spectrum features through the 3D deformable convolution blocks, so that the problem that more useful information cannot be extracted due to the fixation of convolution kernels or information redundancy caused by excessive irrelevant information is reserved when the inter-spectrum features are extracted in the prior art is overcome, and the classification precision of the ground features in the hyperspectral image is improved.
Secondly, the space branch network constructed by the invention enables the small sample spectral space feature extraction convolutional neural network to pay attention to space features of different scales through the multi-scale space feature extraction module, overcomes the defect that the space features of a hyperspectral image block are extracted by using a single scale in the prior art, pays attention to and screens the multi-scale space features through the multi-path space attention mechanism module, extracts more resolved space features, overcomes the information loss caused by insufficient utilization rate of the extracted features in the space feature extraction process in the prior art or retains information redundancy caused by excessive irrelevant information, and improves the classification capability of the convolutional neural network in the training of few samples.
Thirdly, the small sample spectral space feature extraction convolutional neural network adopts a domain attention module, and mainly aims at the problem that redundant information among wave bands is too much due to more spectral wave bands of hyperspectral images. The useful inter-spectral features and the useful spatial features are extracted by spectral attention, so that the neural network focuses more on useful information in the feature information. In order to reduce the domain transfer problem caused by training with different hyperspectral datasets, the invention uses a cross entropy loss function, a correlation alignment loss function and a maximum averaged difference loss function as the loss functions of the network. The convolutional neural network for small sample spectral space feature extraction is more focused on the ground object category with non-centralized sample distribution or small sample quantity.
The effects of the present invention will be further described with reference to simulation experiments.
Simulation experiment conditions:
the hardware platform of the simulation experiment of the invention is: intercore i7-6700, frequency 3.4GHz,Nvidia GeForce RTX3090. The software of the simulation experiment of the present invention uses pytorch.
The simulation experiment of the invention adopts the method and two existing RN-FSC and DCFSL methods to respectively classify the object targets in the University of Pavia and Indian pins hyperspectral data sets.
The RN-FSC method is as follows: a hyperspectral classification method is proposed by Kuiliang Gao et al in Deep relation network for hyperspectral image few-shot classification (Remote Sensing, 2020), and the hyperspectral classification method is characterized in that a characteristic learning module and a relation learning module can fully utilize space-spectrum information in hyperspectral images to realize accurate classification of new hyperspectral images by a small number of marked samples.
The DCFSL method refers to: one hyperspectral classification method that utilizes source class data to aid in the learning of small sample elements for classifying target classes is proposed by Rui Li et al in "Deep crossdomain few-shot learning for hyperspectral image classification" (Remote Sensing, 2021).
The target domain data sets used in the present invention are University of Pavia and Indian pins hyperspectral data sets, which are data collected by the aviis sensor at University of Pavia in california and Indian pine in indiana, usa, respectively. Indian pins was the earliest test data for hyperspectral image classification, an on-board visible infrared imaging spectrometer (aviis) imaging a piece of Indian pine tree in indiana in 1992, and then labeling with a size of 145 x 145 was cut out for hyperspectral image classification test use. The image of University of Pavia hyperspectral dataset is 610×340 in size and has 103 wave bands, and comprises 9 types of features, and the types and the number of each type of feature are shown in table 1.
Table 1University of Pavia sample class and quantity
Figure BDA0004085658520000171
Figure BDA0004085658520000181
The Indian pins hyperspectral dataset image has a size of 145×145 and 200 wave bands, and contains 16 types of features, and the types and the number of each type of feature are shown in table 2.
TABLE 2Indian pins sample categories and amounts
Class label Ground object category Quantity of
1 Alfalfa 46
2 Corn-notill 1428
3 Corn-mintill 830
4 Corn 237
5 Grass-pasture 483
6 Grass-tree 730
7 Grass-pasture-mowed 28
8 Hay-windrowed 478
9 Oats 20
10 Soybean-notill 972
11 Soybean-mintill 2455
12 Soybean-clean 593
13 Wheat 205
14 Woods 1265
15 Buildings-Grass-Trees-Dribes 386
16 Stone-steel-Towers 93
The source domain data set used in the invention is a Chikusei data set, and the Japanese Zhengshi tuxi hyperspectral image is obtained through a Hyperspec-VNIR-CIRIS spectrometer. The ground sampling distance is 2.5m, the image size is 2517×2335 pixels, 512 wavebands are provided, 128 wavebands exist in the spectral range of 363nm to 1018nm, and 19 categories are included in total. The categories and amounts of each type of features are shown in table 3.
Figure BDA0004085658520000182
Figure BDA0004085658520000191
In order to verify the high efficiency and good classification performance of the invention, three evaluation indexes of overall classification precision OA, average precision AA and Kappa coefficient are adopted.
The overall classification accuracy OA refers to the ratio of the number of correctly classified pixels divided by the total number of pixels on the test set, and the value of the overall classification accuracy OA is between 0 and 100%, and the larger the value is, the better the classification effect is.
The average accuracy AA refers to dividing the number of correctly classified pixels of each class on a test set by the total number of all pixels of the class to obtain the correct classification accuracy of the class, and averaging the accuracy of all classes to obtain an average value of 0-100%, wherein the larger the value is, the better the classification effect is.
The Kappa coefficient is an evaluation index defined on the confusion matrix, comprehensively considers elements on the diagonal line of the confusion matrix and elements deviating from the diagonal line, more objectively reflects the classification performance of the algorithm, has a value between-1 and 1, and indicates that the classification effect is better when the value is larger.
2. Simulation experiment content and result analysis:
simulation 1, the classification test is carried out on the invention and the two prior arts in University of Pavia hyperspectral data sets respectively, and the result diagram is shown in fig. 5, wherein:
FIG. 5 (a) is a classification result on a University of Pavia hyperspectral dataset using the existing RN-FSC method;
FIG. 5 (b) is a classification result on a University of Pavia hyperspectral dataset using the existing DCFSL method;
FIG. 5 (c) is a classification result on a University of Pavia hyperspectral dataset using the method of the present invention.
As can be seen from fig. 5 (c), the classification result graph of the present invention on the University of Pavia dataset is significantly smoother and more sharp-edged than fig. 5 (a), fig. 5 (b).
Simulation 2, tested on Indian pins hyperspectral dataset with the present invention and two prior art techniques, respectively, the simulation results are shown in fig. 6, wherein:
FIG. 6 (a) is a classification result on the Indian pins hyperspectral dataset using the existing RN-FSC method;
FIG. 6 (b) is a classification result on the Indian pins hyperspectral dataset using the existing DCFSL method;
FIG. 6 (c) is a classification result on the Indian pins hyperspectral dataset using the method of the present invention;
the results of comparing the accuracy of classification of the present invention and the prior art in University of Pavia and Indian pins hyperspectral datasets, respectively, in the two simulations are shown in table 4.
Table 4 comparison of classification accuracy for three networks under two different data sets
Figure BDA0004085658520000201
Figure BDA0004085658520000211
As can be seen from Table 4, the method of the invention obtains higher classification accuracy than the RN-FSC method and the DCFSL method in the prior art under University of Pavia and Indian pins data sets, which shows that the method can more accurately predict the types of hyperspectral image samples.
The simulation experiment shows that: the method can more fully extract the inter-spectrum features by using the constructed inter-spectrum 3D deformable convolution block, and the constructed multi-scale space feature extraction block can more fully extract the space features. And splicing the spatial features and the inter-spectrum features, obtaining spectral-spatial features with more distinguishability through a full-connection layer, and finally obtaining a hyperspectral image classification result through a softmax classifier. The invention trains the neural network by adopting the total loss function formed by the cross entropy loss function, the correlation alignment loss function and the maximum average difference loss function, so that the small sample spectral space characteristic extraction convolution neural network is more concerned with the ground object category with non-concentrated sample distribution or small sample quantity. The method solves the problem that the classification accuracy is not high under the condition of few training samples due to information loss caused by insufficient utilization rate of the extracted features or information redundancy caused by excessive irrelevant information retention in the prior art during space feature extraction, and is a very practical hyperspectral image classification method aiming at the few training samples.
In the description of the invention, the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
In the description of the present specification, a description of the terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic data point described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristic data points described may be combined in any suitable manner in any one or more embodiments or examples. Further, one skilled in the art can engage and combine the different embodiments or examples described in this specification. The foregoing is a further detailed description of the invention in connection with the preferred embodiments, and it is not intended that the invention be limited to the specific embodiments described. It will be apparent to those skilled in the art that several simple deductions or substitutions may be made without departing from the spirit of the invention, and these should be considered to be within the scope of the invention.

Claims (8)

1. The hyperspectral image classification method based on the deep learning of the small sample is characterized by comprising the following steps of:
step 1, acquiring a source domain hyperspectral image set and a target domain hyperspectral image set, wherein the source domain hyperspectral image set comprises a plurality of Zhang Yuanyu hyperspectral images, and the target domain hyperspectral image set comprises a plurality of target domain hyperspectral images;
step 2, respectively filling the edge parts of the source domain hyperspectral image and the target domain hyperspectral image, and extracting a plurality of first image blocks and a plurality of second image blocks by taking each pixel point of the filled source domain hyperspectral image and target domain hyperspectral image as a center;
step 3, in each category, randomly selecting part of the first image blocks to form a source domain support set and part of the first image blocks to form a source domain query set, and randomly selecting part of the second image blocks to form a target domain support set and part of the first image blocks to form a target domain query set;
step 4, based on a random gradient descent method, carrying out alternate training on a small sample spectral space feature extraction convolutional neural network by using the source domain support set, the source domain query set, the target domain support set and the target domain query set to obtain a trained small sample spectral space feature extraction convolutional neural network, wherein the small sample spectral space feature extraction convolutional neural network extracts features from two aspects of spectrum and space, and a total loss function of the small sample spectral space feature extraction convolutional neural network consists of a cross entropy loss function, a correlation alignment loss function and a maximum average difference loss function;
And step 5, inputting the hyperspectral image to be classified into the trained small sample spectral space feature extraction convolutional neural network to obtain a classification result.
2. The hyperspectral image classification method based on small sample deep learning as claimed in claim 1, wherein the step 2 includes:
step 2.1, filling pixels with pixel values of 0 into the peripheries of the source domain hyperspectral image and the target domain hyperspectral image respectively to obtain a filled source domain hyperspectral image and a filled target domain hyperspectral image;
and 2.2, selecting the first image block and the second image block with the space size of (2t+1) x (2t+1) and the channel number of d by taking each pixel point in the filled source domain hyperspectral image and the filled target domain hyperspectral image as a center, wherein t is an integer greater than 0.
3. The hyperspectral image classification method based on small sample deep learning as claimed in claim 1, wherein the structure of the small sample spectral space feature extraction convolutional neural network comprises a spectral branch network, a spatial branch network, a domain attention module, a first splicing layer, a full connection layer and a softmax classifier, the spectral branch network, the spatial branch network and the domain attention module are connected in parallel and then sequentially connected in series, the domain attention module, the first splicing layer, the full connection layer and the softmax classifier are connected in series, the spectral branch network comprises 2 3D deformable convolution blocks and 2 first maximum pooling layers, the 1 st 3D deformable convolution block, the 1 st first maximum pooling layer, the 2 nd 3D deformable convolution block and the 2 nd first maximum pooling layer are sequentially connected in series, and the spatial branch network domain attention module comprises a multi-scale spatial feature extraction module, a second splicing layer and a second maximum pooling layer which are sequentially connected in series.
4. The hyperspectral image classification method based on small sample deep learning as claimed in claim 3, wherein the 3D deformable convolution block includes 3D deformable convolution layers and 3 first activation function layers, the 1 st 3D deformable convolution layer, the 1 st first activation function layer, the 2 nd 3D deformable convolution layer, the 2 nd first activation function layer, the 3 rd 3D deformable convolution layer, and the 3 rd first activation function layer are sequentially connected in series, and the output of the 1 st first activation function layer and the output of the 3 rd 3D deformable convolution layer are added to form a residual structure.
5. The hyperspectral image classification method based on small sample deep learning as claimed in claim 3, wherein the multi-scale spatial feature extraction module comprises 2 scale operation layers, 3 convolution layers, 3 normalization layers and 3 second activation function layers;
the 1 st convolution layer, the 1 st normalization layer and the 1 st second activation function layer are sequentially connected in series;
the 1 st scale operation layer, the 2 nd convolution layer, the 2 nd normalization layer and the 2 nd second activation function layer are sequentially connected in series;
the 2 nd scale operation layer, the 3 rd convolution layer, the 3 rd normalization layer and the 3 rd second activation function layer are sequentially connected in series;
The 1 st second activation function layer, the 2 nd second activation function layer and the 3 rd second activation function layer are connected in parallel and then connected in series with the second splicing layer and the second maximum pooling layer in sequence.
6. The hyperspectral image classification method based on small sample deep learning as claimed in claim 1, wherein the total loss function of the small sample spectral null feature extraction convolutional neural network is:
L total ＝L fsl +L coral +L MMD
wherein ,Ltotal Representing the total loss function L of the small sample spectral space feature extraction convolutional neural network fsl Represents cross entropy loss, L coral Representing dependency alignment loss, L MMD Representing the maximum averaged delta loss;
cross entropy loss L fsl Expressed as:
Figure FDA0004085658510000031
Figure FDA0004085658510000032
wherein d (·) represents the Euclidean distance, F ω (. Cndot.) the feature extraction function of parameter ω, f l Features representing the first class in the source domain support set or the target domain support set, C being the number of classes, x j Representing one sample in a source domain query set or a target domain query set, y j Representing sample x j Q represents a source domain query set or a target domain query set;
correlation alignment loss L coral Expressed as:
Figure FDA0004085658510000041
wherein ,
Figure FDA0004085658510000046
frobenius norms, C representing matrix S Covariance matrix representing source domain features, C T A covariance matrix representing the characteristics of the target domain, and d represents the dimension of the characteristics;
maximum averaged difference loss L MMD Expressed as:
Figure FDA0004085658510000042
wherein ,
Figure FDA0004085658510000043
represents the spatial distance, phi (·) represents the mapping function, +.>
Figure FDA0004085658510000044
Representing source domain features, ++>
Figure FDA0004085658510000045
Representing the characteristics of the target domain, X s Representing a source domain dataset, X t Representing a target domain dataset, n s Representing the number of data in the source domain dataset, n t Representing the number of data in the target domain dataset.
7. The hyperspectral image classification method based on small sample deep learning as claimed in claim 6, wherein the step 4 includes:
setting the initial learning rate of training as alpha and the iteration times as T, and sending the source domain support set and the source domain query set into the small sample spectral space feature extraction convolutional neural network for training when the iteration is performed for odd times, and calculating the loss value between the features of the source domain support set and the features of the source domain query set by using the total loss function so as to update the parameters in the small sample spectral space feature extraction convolutional neural network; and when the number of iterations is even, the target domain support set and the target domain query set are sent into the small sample spectral space feature extraction convolutional neural network to train, the total loss function is utilized to calculate the loss value between the features of the target domain support set and the features of the target domain query set so as to update the parameters in the small sample spectral space feature extraction convolutional neural network until the loss value of the small sample spectral space feature extraction convolutional neural network is not reduced any more and the current training round number is smaller than the iteration number T or the training round number reaches the iteration number T, and training of the small sample spectral space feature extraction convolutional neural network is stopped to obtain the trained small sample spectral space feature extraction convolutional neural network.
8. The hyperspectral image classification method based on small sample deep learning as claimed in claim 7, wherein the small sample spectral null feature extraction convolutional neural network updated weight vector W new The method comprises the following steps:
Figure FDA0004085658510000051
wherein ,Ltotal The method is characterized in that the method comprises the steps of representing a total loss function of a small sample spectral null feature extraction convolutional neural network, W represents a weight vector before updating the small sample spectral null feature extraction convolutional neural network, and R represents a learning rate.
CN202310095139.8A 2023-02-08 2023-02-08 Hyperspectral image classification method based on small sample deep learning Pending CN116310510A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310095139.8A CN116310510A (en) 2023-02-08 2023-02-08 Hyperspectral image classification method based on small sample deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310095139.8A CN116310510A (en) 2023-02-08 2023-02-08 Hyperspectral image classification method based on small sample deep learning

Publications (1)

Publication Number Publication Date
CN116310510A true CN116310510A (en) 2023-06-23

Family

ID=86826583

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310095139.8A Pending CN116310510A (en) 2023-02-08 2023-02-08 Hyperspectral image classification method based on small sample deep learning

Country Status (1)

Country Link
CN (1) CN116310510A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116977747A (en) * 2023-08-28 2023-10-31 中国地质大学（北京） Small sample hyperspectral classification method based on multipath multi-scale feature twin network
CN117132804A (en) * 2023-07-04 2023-11-28 中国矿业大学 Hyperspectral image classification method based on causal cross-domain small sample learning
CN117474928A (en) * 2023-12-28 2024-01-30 东北大学 Ceramic package substrate surface defect detection method based on meta-learning model
CN117809293A (en) * 2024-03-01 2024-04-02 电子科技大学 Small sample image target counting method based on deep neural network
CN118411577A (en) * 2024-07-04 2024-07-30 中国人民解放军国防科技大学 Cross-domain small sample multispectral image classification method, device and computer equipment

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117132804A (en) * 2023-07-04 2023-11-28 中国矿业大学 Hyperspectral image classification method based on causal cross-domain small sample learning
CN117132804B (en) * 2023-07-04 2024-04-05 中国矿业大学 Hyperspectral image classification method based on causal cross-domain small sample learning
CN116977747A (en) * 2023-08-28 2023-10-31 中国地质大学（北京） Small sample hyperspectral classification method based on multipath multi-scale feature twin network
CN116977747B (en) * 2023-08-28 2024-01-23 中国地质大学（北京） Small sample hyperspectral classification method based on multipath multi-scale feature twin network
CN117474928A (en) * 2023-12-28 2024-01-30 东北大学 Ceramic package substrate surface defect detection method based on meta-learning model
CN117474928B (en) * 2023-12-28 2024-03-19 东北大学 Ceramic package substrate surface defect detection method based on meta-learning model
CN117809293A (en) * 2024-03-01 2024-04-02 电子科技大学 Small sample image target counting method based on deep neural network
CN117809293B (en) * 2024-03-01 2024-05-03 电子科技大学 Small sample image target counting method based on deep neural network
CN118411577A (en) * 2024-07-04 2024-07-30 中国人民解放军国防科技大学 Cross-domain small sample multispectral image classification method, device and computer equipment
CN118411577B (en) * 2024-07-04 2024-09-24 中国人民解放军国防科技大学 Cross-domain small sample multispectral image classification method, device and computer equipment

Similar Documents

Publication Publication Date Title
CN116310510A (en) Hyperspectral image classification method based on small sample deep learning
CN112052755B (en) Semantic convolution hyperspectral image classification method based on multipath attention mechanism
CN113095409B (en) Hyperspectral image classification method based on attention mechanism and weight sharing
CN109117956B (en) Method for determining optimal feature subset
CN107992891B (en) Multispectral remote sensing image change detection method based on spectral vector analysis
CN111639587B (en) Hyperspectral image classification method based on multi-scale spectrum space convolution neural network
CN104463203B (en) High-spectrum remote sensing semisupervised classification method based on the scoring of atural object classification degree of membership
CN108280396A (en) Hyperspectral image classification method based on depth multiple features active migration network
CN111160396B (en) Hyperspectral image classification method of graph convolution network based on multi-graph structure
CN104392454B (en) The merging method based on the scoring of atural object classification degree of membership under the empty spectrum combining classification framework of high-spectrum remote sensing
CN108229551B (en) Hyperspectral remote sensing image classification method based on compact dictionary sparse representation
CN104820841B (en) Hyperspectral classification method based on low order mutual information and spectrum context waveband selection
CN104820840B (en) The arest neighbors hyperspectral image classification method recombinated based on dictionary and wave band
CN109583469B (en) K-means hyperspectral image band clustering method based on mutual information
CN113936214B (en) Karst wetland vegetation community classification method based on fusion of aerospace remote sensing images
CN110543916B (en) Method and system for classifying missing multi-view data
JP6945253B2 (en) Classification device, classification method, program, and information recording medium
CN111126361B (en) SAR target identification method based on semi-supervised learning and feature constraint
CN111222545B (en) Image classification method based on linear programming incremental learning
CN112396619A (en) Small particle segmentation method based on semantic segmentation and internally complex composition
CN111680579B (en) Remote sensing image classification method for self-adaptive weight multi-view measurement learning
CN112749621A (en) Remote sensing image cloud layer detection method based on deep convolutional neural network
CN114299398B (en) Small sample remote sensing image classification method based on self-supervision contrast learning
CN113469084B (en) Hyperspectral image classification method based on contrast generation countermeasure network
CN104732246B (en) A kind of semi-supervised coorinated training hyperspectral image classification method

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