CN113449603B - High-resolution remote sensing image earth surface element identification method and storage medium - Google Patents
High-resolution remote sensing image earth surface element identification method and storage medium Download PDFInfo
- Publication number
- CN113449603B CN113449603B CN202110594123.2A CN202110594123A CN113449603B CN 113449603 B CN113449603 B CN 113449603B CN 202110594123 A CN202110594123 A CN 202110594123A CN 113449603 B CN113449603 B CN 113449603B
- Authority
- CN
- China
- Prior art keywords
- features
- remote sensing
- level
- convolution
- surface elements
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a method for identifying earth surface elements of a high-resolution remote sensing image and a storage medium, wherein the method for identifying the earth surface elements comprises the following steps: step 1: acquiring low-level space spectrum features; and 2, step: fusing the low-level spatial spectrum features through a shallow spatial spectrum feature fusion model; and 3, step 3: fusing the fused features in the step 2 through a middle-layer multi-scale feature fusion model; and 4, step 4: fusing the fused features in the step 3 through a deep multi-level feature fusion model; and 5: obtaining high-level semantic features according to the fusion features output in the step 4; step 6: and classifying the surface elements through a classifier to obtain a surface element identification result. Compared with the prior art, the method has the advantages of high precision, high speed, accuracy, rapidness and the like.
Description
Technical Field
The invention relates to the technical field of high-resolution remote sensing image earth surface element identification, in particular to a high-resolution remote sensing image earth surface element identification method based on space spectrum feature fusion and a storage medium.
Background
The feature extraction is a key step of multi-feature fusion and earth surface element identification of traditional machine learning and deep learning, and a key technology of remote sensing image feature fusion and processing analysis is provided for how to extract features reflecting different modes. With the increasing depth of research, the way of feature extraction is continuously extended, from unsupervised to (semi-) supervised, from spectrum or space to spectrum and space, from manual to automatic, from manual to end-to-end, from shallow to deep. For high-resolution, especially ultra-high-resolution remote sensing image data, the contour details of the surface elements are high, the spectrum mixed elements corresponding to the surface covering elements are various, and particularly, the fine discrimination difficulty of sub-class elements (such as grasslands, trees, farmlands under vegetation, and even corn, vegetable lands, wheat lands and the like which can be divided under the farmlands) is high, so that the phenomenon of 'same-object different-spectrum' or 'same-spectrum foreign matter' is more prominent. In order to avoid the deficiency of identifying surface elements by using single-source features, various researchers have sought to improve the accuracy of identifying surface elements by using multi-feature fusion. However, the existing fusion method is usually based on multi-feature direct superposition, which results in increased feature dimensions and high redundancy after superposition, and is difficult to accurately depict the discriminant features of the earth surface elements, and cannot well realize the joint expression of multi-feature significant information, so that the high-precision identification of earth surface coverage cannot be effectively realized.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a high-precision, high-speed, accurate and quick high-resolution remote sensing image earth surface element identification method and a storage medium.
The purpose of the invention can be realized by the following technical scheme:
a high-resolution remote sensing image ground surface element identification method comprises the following steps:
step 1: acquiring low-level space spectrum features;
and 2, step: fusing the low-level spatial spectrum features through a shallow spatial spectrum feature fusion model;
and 3, step 3: fusing the fused features in the step 2 through a middle-layer multi-scale feature fusion model;
and 4, step 4: fusing the fused features in the step 3 through a deep multi-level feature fusion model;
and 5: obtaining high-level semantic features according to the fusion features output in the step 4;
and 6: and classifying the surface elements through a classifier to obtain a surface element identification result.
Preferably, the low-level spatial spectral features in step 1 are obtained by an existing spatial spectral feature extraction method.
Preferably, the step 2 specifically comprises:
first, perform depth separable convolution, separatableconv, on the input low-level spatial spectral features;
secondly, down-sampling by adopting two pooling layers of maximum pooling MaxPool and average pooling AvgPool, and directly cascading the pooling result with Concat;
and finally, calibrating the information contribution degree of the acquired multiple convolution characteristics by using an attention mechanism module SE.
Preferably, the step 3 specifically comprises: and simultaneously outputting the features after the shallow layer space spectrum fusion to three convolution feature extraction modules, and fusing convolution features of different receptive fields by utilizing a cascade Concat and attention mechanism module SE.
More preferably, the three convolution feature extraction modules are specifically:
the convolution feature extraction modules with the receptive fields of 1 × 1, 3 × 3 and 5 × 5 are respectively used for learning spatial context information of different scales; the 3 x 3 convolution feature extraction module is used for learning spatial correlation, and the 1 x 1 convolution feature extraction module is used for learning correlation among channels.
Preferably, the step 4 specifically includes:
firstly, performing 3 × 3 convolution and maximum pooling for a plurality of times, and gradually performing down-sampling;
secondly, flattening the acquired different-layer features by using Flatten and full-connection Dense, and then cascading Concat;
and finally, calibrating the information contribution degree of the acquired multi-level high-level semantic features by using an attention mechanism module SE.
More preferably, the step 4 performs convolution with 3 × 3 and maximal pooling three times, and performs flattening Flatten and full join density on the obtained features once, and finally performs cascading Concat on the features of each layer.
Preferably, the step 5 specifically comprises:
and (4) processing the fusion features output in the step (4) by adopting a full connection layer to obtain high-level semantic features.
Preferably, the classifier in the step 6 is a Softmax function layer.
A storage medium is provided, wherein any one of the methods for identifying the earth surface elements of the high-resolution remote sensing image is stored in the storage medium.
Compared with the prior art, the invention has the following beneficial effects:
1. and (3) realizing high-precision classification: the high-resolution remote sensing image identification method introduces an SE attention mechanism, can redistribute the contribution degrees of fusion features of different levels and different scales according to the importance degree, and realizes the accurate identification and high-precision classification of typical earth surface elements of the high-resolution remote sensing image through the feature fusion of different scales, different levels and different semantics.
2. The processing speed is high: the high-resolution remote sensing image identification method combines multi-feature fusion and deep learning together to identify the earth surface elements from end to end, and has the advantages of strong semantic information, high automation degree, strong adaptability and high data processing speed.
Drawings
FIG. 1 is a schematic flow chart of a method for identifying surface elements of high-resolution remote sensing images in the invention;
FIG. 2 is a pseudo color composite map and ground reference map of high resolution Quickbird data for a ZH scene in an embodiment of the invention;
wherein, FIG. 2 (a) is a pseudo color composition diagram of a ZH scene, and FIG. 2 (b) is a ground reference diagram of a ZH scene;
FIG. 3 is a diagram illustrating overall accuracy variations after ten trainings performed by different methods according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of classification diagrams and overall classification accuracy of different methods according to an embodiment of the present invention;
wherein, fig. 4 (a) represents the classification result of the SVM method, fig. 4 (b) represents the classification result of the RF method, fig. 4 (c) represents the classification result of the VGG method, fig. 4 (d) represents the classification result of the FDSSC method, fig. 4 (e) represents the classification result of the SSRN method, fig. 4 (f) represents the classification result of the SSUN method, and fig. 4 (g) represents the classification result of the SSFN method according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, shall fall within the protection scope of the present invention.
As shown in fig. 1, the present embodiment relates to a method for identifying surface elements of a high-resolution remote sensing image based on a Spectral-Spatial Fusion Network (SSFN), which is used for acquiring discriminative Fusion features of different scales, different levels, and different semantics through four end-to-end series learning steps and implementing high-precision identification and classification of typical surface elements on a high-resolution remote sensing image, and specifically includes:
step 1: acquiring low-level space spectrum features by using an existing space spectrum feature extraction method in the prior art;
and 2, step: fusing the low-level spatial spectrum features through a shallow spatial spectrum feature fusion model;
first, perform depth separable convolution, separatableconv, on the input low-level spatial spectral features;
secondly, down-sampling by adopting two pooling layers of maximum pooling MaxPool and average pooling AvgPool, and directly cascading the pooling result with Concat;
finally, calibrating the information contribution degree of the acquired multiple convolution characteristics by using an attention mechanism module SE;
and step 3: fusing the fused features in the step 2 through a middle-layer multi-scale feature fusion model;
and simultaneously inputting the features after the shallow space spectrum fusion to convolution feature extraction modules of three different receptive fields, and performing cascade fusion. Firstly, three receptive fields of 1 × 1, 3 × 3 and 5 × 5 are formed by combining different numbers of 3 × 3 convolution feature extraction modules and 1 × 1 convolution feature extraction modules, and are used for learning spatial context information of different scales. Wherein, the convolution of 3 × 3 is used for learning spatial correlation, and the convolution of 1 × 1 is used for learning correlation between channels; then, fusing convolution characteristics of different receptive fields by utilizing a cascade Concat and attention mechanism module SE;
and 4, step 4: fusing the fused features in the step 3 through a deep multi-level feature fusion model;
considering that the detail information of the ground features of the high-resolution remote sensing image is rich, the local spectrum difference of the ground features is large, and more salt and pepper noise points may exist when the pixel-by-pixel classification is directly carried out. Therefore, after the middle layer fusion features are extracted, firstly, 3 × 3 convolution and maximum pooling are performed, and downsampling is performed step by step to reduce the difference of the local spectra of the ground features;
secondly, performing flattening and full connection of the acquired features of different levels, and then cascading, so that the semantic expression capability of the features is stronger while different levels of detail information are kept;
finally, an SE attention mechanism module is used for calibrating the information contribution degree of the acquired multilayer deep semantic features, so that the significance expression of different semantic features is further improved;
in this step, the convolution with 3 × 3 and the maximum pooling are performed three times, and each time the convolution is performed, the flattening Flatten and the full-link density are performed on the obtained features, and finally the cascading Concat is performed on the features of each layer.
And 5: processing the fusion features output in the step 4 by adopting a full connection layer to obtain high-level semantic features;
and 6: and classifying the surface elements through a classifier to obtain a surface element identification result.
The embodiment also relates to a storage medium, wherein any one of the high-resolution remote sensing image surface feature identification methods is stored in the storage medium.
In the embodiment, a high-resolution Quickbird satellite remote sensing image of a ZH scene is selected, and high-resolution data adopted by an experiment is a panchromatic sharpening result with the resolution close to 0.62 m, which is generated by fusing original multispectral wave bands (blue, green, red and near-infrared wave bands) and panchromatic wave bands. Fig. 2 shows a pseudo color composite of the ZH panchromatic sharpening data set fig. 2 (a) and the reference fig. 2 (b). The ZH data set contains 4 surface element classes of road, building, tree and grass. Table 1 provides detailed information of each table element reference sample of the ZH dataset.
TABLE 1 Total sample of ZH data set and number of experimental training samples
| Sample classes | Total sample (Pixel) | Training sample (Pixel) |
| Road | 86551 | 865 |
| Construction of buildings | 154164 | 1541 |
| Tree | 81033 | 811 |
| Grass land | 7201 | 72 |
The experimental results are as follows:
in order to provide richer low-level visual features for the input of different surface element identification methods, on the basis, an extended morphological attribute profile of ZH data is extracted experimentally as a typical spatial spectrum multi-feature input. The specific evaluation indexes are selected from four types of Overall precision (OA for short), kappa Coefficient (Kappa Coefficient for short), single type precision (CA for short) and Time consumption (Time T for short).
The performance and stability of the different methods can be reflected by the differences between the mean, maximum and minimum values of the overall accuracy under ten training sessions. Fig. 3 shows the average, maximum and minimum values of the overall accuracy under ten training sessions of different methods. It can be seen that the highest precision is the SSFN method proposed in this embodiment, and the highest precision reaches 98.41%. Secondly, the precision of a popular space spectrum feature fusion network FDSSC is as high as 98.33 percent, and the highest precision of a classical convolutional neural network VGG is 98.05 percent. The worst but also the most stable precision is the SVM and RF with traditional machine learning support, with overall precision of 91.95% and 94.63%, respectively. Compared with the deep learning method with high precision, the SSFN method proposed by the present embodiment is relatively stable and has the highest precision.
Table 2 shows the average accuracy evaluation results of ten times of training in different methods. It can be seen that the average result of the overall accuracy is consistent with the analysis of fig. 3, wherein the highest accuracy is the SSFN method proposed by the present invention, and the average of the overall accuracy is 98.11%. Next, FDSSC, VGG, and SSRN methods, with overall accuracy averages of 97.83%, 97.88%, and 96.84%, respectively. From the comparison of the single precision of the 'grassland' surface elements, the 'grassland' precision of the SSFN is 83.17 percent, which is much higher than that of other methods, and the SSFN is proved to obtain the distinguishing characteristics which are more favorable for expressing different surface elements, particularly the easily confused surface elements through effective characteristic fusion, so that the distinguishing degree of the surface elements is improved, and the integral identification precision of the surface elements is further improved. Finally, comparing the time consumption T of different methods, the time consumption (152.62 seconds) of the SSFN method under the deep learning support is far lower than the three methods of FDSSC (839.42 seconds), VGG (207.84 seconds) and SSRN (414.80 seconds) with higher accuracy under the deep learning support, and it is further demonstrated that the proposed SSFN method is higher in efficiency and better in performance.
FIG. 4 gives classification plots for different methods of ZH data set and overall classification accuracy. Comparing the ground reference diagram shown in fig. 2 (b), it can be seen that the classification accuracy of the conventional machine learning model SVM and RF is the lowest (91.95% and 94.63%), and the classification result is the worst. Because a single pixel vector is directly selected as a classification input, the significance and the spatial context information of the ground features cannot be well integrated, so that more wrongly-divided pixels exist, and the salt and pepper phenomenon is very obvious. Compared with the classification results of the classical network VGG and the flow space spectrum feature fusion networks fdsscc, SSRN and SSUN supported by deep learning, the SSFN provided by the embodiment has the highest overall precision (98.41%), the best classification effect and relatively less local area pixel mismatching (see fig. 4 (g)).
In conclusion, on the ZH data set, through a series of qualitative and quantitative experimental analyses, the result proves that the proposed novel space spectrum feature fusion network SSFN can realize higher-precision ground surface element identification research compared with other methods, and has very obvious advantages in accurately describing the geometric boundaries of the ground surface elements and the uniformity of internal spectra.
TABLE 2 results of precision evaluation by different methods
While the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (7)
1. A high-resolution remote sensing image ground surface element identification method is characterized by comprising the following steps:
step 1: acquiring low-level space spectrum features;
step 2: fusing the low-level spatial spectrum features through a shallow spatial spectrum feature fusion model;
the step 2 specifically comprises the following steps:
first, perform depth separable convolution, separatableconv, on the input low-level spatial spectral features;
secondly, down-sampling by adopting two pooling layers of maximum pooling MaxPool and average pooling AvgPool, and directly cascading the pooling result with Concat;
finally, calibrating the information contribution degree of the acquired various convolution characteristics by using an attention mechanism module SE;
and step 3: fusing the fused features in the step 2 through a middle-layer multi-scale feature fusion model;
the step 3 specifically comprises the following steps: the features after the shallow layer space spectrum fusion are simultaneously output to three convolution feature extraction modules, and convolution features of different receptive fields are fused by utilizing a cascade Concat and attention mechanism module SE;
and 4, step 4: fusing the fused features in the step 3 through a deep multi-level feature fusion model;
the step 4 is specifically as follows:
firstly, performing 3 × 3 convolution and maximum pooling for a plurality of times, and gradually performing down-sampling;
secondly, flattening the acquired different-layer features by using Flatten and full-connection Dense, and then cascading Concat;
finally, calibrating the information contribution degree of the acquired multi-level high-level semantic features by using an attention mechanism module SE;
and 5: obtaining high-level semantic features according to the fusion features output in the step 4;
step 6: and classifying the surface elements through a classifier to obtain a surface element identification result.
2. The method for identifying the earth surface elements of the high-resolution remote sensing images according to claim 1, wherein the low-level spatial spectral features in the step 1 are obtained by an existing spatial spectral feature extraction method.
3. The method for identifying the earth surface elements of the high-resolution remote sensing image according to claim 1, wherein the three convolution feature extraction modules are specifically:
the convolution feature extraction modules with the receptive fields of 1 × 1, 3 × 3 and 5 × 5 are respectively used for learning spatial context information of different scales; the 3 x 3 convolution feature extraction module is used for learning spatial correlation, and the 1 x 1 convolution feature extraction module is used for learning correlation among channels.
4. The method for identifying the earth surface elements of the high-resolution remote sensing images according to claim 1, wherein step 4 is implemented by performing three times of 3 x 3 convolution and maximum pooling, performing flattening Flatten and full-link Dense on the acquired features once, and finally performing cascading Concat on the features of all layers.
5. The method for identifying the earth surface elements of the high-resolution remote sensing images according to claim 1, wherein the step 5 specifically comprises the following steps:
and (4) processing the fusion features output in the step (4) by adopting a full connection layer to obtain high-level semantic features.
6. The method for identifying the earth surface elements in the high-resolution remote sensing images as claimed in claim 1, wherein the classifier in the step 6 is a Softmax function layer.
7. A storage medium, wherein the method for identifying the surface elements of the high-resolution remote sensing images according to any one of claims 1 to 6 is stored in the storage medium.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110594123.2A CN113449603B (en) | 2021-05-28 | 2021-05-28 | High-resolution remote sensing image earth surface element identification method and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110594123.2A CN113449603B (en) | 2021-05-28 | 2021-05-28 | High-resolution remote sensing image earth surface element identification method and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113449603A CN113449603A (en) | 2021-09-28 |
| CN113449603B true CN113449603B (en) | 2022-10-25 |
Family
ID=77810344
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110594123.2A Active CN113449603B (en) | 2021-05-28 | 2021-05-28 | High-resolution remote sensing image earth surface element identification method and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113449603B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113673487A (en) * | 2021-10-21 | 2021-11-19 | 杭州觅睿科技股份有限公司 | Pet identification method, device, equipment and storage medium |
| CN115082778B (en) * | 2022-04-28 | 2023-04-07 | 中国农业科学院农业信息研究所 | Multi-branch learning-based homestead identification method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112733659A (en) * | 2020-12-30 | 2021-04-30 | 华东师范大学 | Hyperspectral image classification method based on self-learning double-flow multi-scale dense connection network |
| CN112836666A (en) * | 2021-02-19 | 2021-05-25 | 河海大学 | Hyperspectral image classification and identification method |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107837082B (en) * | 2017-11-27 | 2020-04-24 | 乐普（北京）医疗器械股份有限公司 | Automatic electrocardiogram analysis method and device based on artificial intelligence self-learning |
| CN110390952B (en) * | 2019-06-21 | 2021-10-22 | 江南大学 | City sound event classification method based on dual-feature 2-DenseNet parallel connection |
| CN111709291B (en) * | 2020-05-18 | 2023-05-26 | 杭州电子科技大学 | Takeaway personnel identity recognition method based on fusion information |
| CN111696101A (en) * | 2020-06-18 | 2020-09-22 | 中国农业大学 | Light-weight solanaceae disease identification method based on SE-Inception |
| CN111914916B (en) * | 2020-07-22 | 2022-08-26 | 河海大学 | Hyperspectral image classification model and method based on double-path convolution fusion neural network |
-
2021
- 2021-05-28 CN CN202110594123.2A patent/CN113449603B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112733659A (en) * | 2020-12-30 | 2021-04-30 | 华东师范大学 | Hyperspectral image classification method based on self-learning double-flow multi-scale dense connection network |
| CN112836666A (en) * | 2021-02-19 | 2021-05-25 | 河海大学 | Hyperspectral image classification and identification method |
Non-Patent Citations (1)
| Title |
|---|
| Spectral-Spatial Unified Network for Hyperspectral Image Classification;Yonghao Xu et al.;《IEEE Transactions on Geoscience and Remote Sensing》;20180509;第56卷(第10期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113449603A (en) | 2021-09-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111914907B (en) | Hyperspectral image classification method based on deep learning space-spectrum combined network | |
| Zhu et al. | A spatial-channel progressive fusion ResNet for remote sensing classification | |
| CN111523521B (en) | Remote sensing image classification method for double-branch fusion multi-scale attention neural network | |
| CN111476170A (en) | Remote sensing image semantic segmentation method combining deep learning and random forest | |
| CN108052966B (en) | Remote sensing image scene automatic extraction and classification method based on convolutional neural network | |
| CN111582194B (en) | Multi-temporal high-resolution remote sensing image building extraction method based on multi-feature LSTM network | |
| CN113449603B (en) | High-resolution remote sensing image earth surface element identification method and storage medium | |
| CN107330875B (en) | Water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing image | |
| CN111652038A (en) | Remote sensing sea ice image classification method based on convolutional neural network | |
| CN102308306A (en) | A constraint generator for use in image segregation | |
| CN112949416B (en) | Supervised hyperspectral multiscale graph volume integral classification method | |
| CN111639587B (en) | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network | |
| CN108121949B (en) | A kind of harbour Ore stockpile recognition methods based on remote sensing scene classification | |
| CN108710864B (en) | Winter wheat remote sensing extraction method based on multi-dimensional identification and image noise reduction processing | |
| CN111008664A (en) | Hyperspectral sea ice detection method based on space-spectrum combined characteristics | |
| CN115512162A (en) | Terrain classification method based on attention twin network and multi-mode fusion features | |
| Zhan et al. | Vegetation land use/land cover extraction from high-resolution satellite images based on adaptive context inference | |
| CN115909052A (en) | Hyperspectral remote sensing image classification method based on hybrid convolutional neural network | |
| Ma et al. | A multi-scale progressive collaborative attention network for remote sensing fusion classification | |
| Lone et al. | Object detection in hyperspectral images | |
| Hasanlou et al. | A sub-pixel multiple change detection approach for hyperspectral imagery | |
| CN111738052A (en) | Multi-feature fusion hyperspectral remote sensing ground object classification method based on deep learning | |
| Wemmert et al. | Multiresolution remote sensing image clustering | |
| CN116452872A (en) | Forest scene tree classification method based on improved deep pavv3+ | |
| CN116091940B (en) | Crop classification and identification method based on high-resolution satellite remote sensing image |
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 |