CN114898097A - Image recognition method and system - Google Patents
Image recognition method and system Download PDFInfo
- Publication number
- CN114898097A CN114898097A CN202210623004.XA CN202210623004A CN114898097A CN 114898097 A CN114898097 A CN 114898097A CN 202210623004 A CN202210623004 A CN 202210623004A CN 114898097 A CN114898097 A CN 114898097A
- Authority
- CN
- China
- Prior art keywords
- image data
- data
- remote sensing
- sentinel
- mangrove forest
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 240000002044 Rhizophora apiculata Species 0.000 claims abstract description 139
- 238000012549 training Methods 0.000 claims abstract description 90
- 238000012545 processing Methods 0.000 claims abstract description 41
- 238000005070 sampling Methods 0.000 claims abstract description 15
- 230000011218 segmentation Effects 0.000 claims description 97
- 238000012795 verification Methods 0.000 claims description 46
- 238000012360 testing method Methods 0.000 claims description 38
- 230000006870 function Effects 0.000 claims description 34
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 23
- 241000120622 Rhizophoraceae Species 0.000 claims description 16
- 238000002372 labelling Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000004519 manufacturing process Methods 0.000 claims description 4
- 230000010365 information processing Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 description 12
- 238000012706 support-vector machine Methods 0.000 description 8
- 238000013136 deep learning model Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 1
- 230000009193 crawling Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005906 menstruation Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/10—Image acquisition
- G06V10/12—Details of acquisition arrangements; Constructional details thereof
- G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
- G06V10/143—Sensing or illuminating at different wavelengths
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—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
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/30—Assessment of water resources
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)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Remote Sensing (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Astronomy & Astrophysics (AREA)
- Computational Linguistics (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides an image identification method and system, which comprises the steps of firstly carrying out up-sampling processing on Sentinel-2 data with 10-meter-level resolution to obtain sub-meter-level image data, carrying out wave band operation on the sub-meter-level image data, and fusing index information obtained after the wave band operation, a plurality of wave band information and wave band information of RGB remote sensing image data with the sub-meter-level resolution to obtain second image data; segmenting the RGB remote sensing image data, removing non-mangrove forest growing areas and carrying out binarization to obtain label image data; performing tiling processing on the RGB remote sensing image data, the second image data and the label image data to construct an initial sample set for training an image recognition model; and finally, predicting the mangrove forest object of the remote sensing image data to be predicted according to the pre-trained image recognition model. The invention can quickly and accurately identify the mangrove forest in the global scope, thereby obtaining the mangrove forest distribution map in the global scope.
Description
Technical Field
The invention relates to the technical field of remote sensing, in particular to an image identification method and system.
Background
The remote sensing image identification is a technology for analyzing spectral information and spatial information of various ground features in a remote sensing image by using a computer and dividing each pixel in the image into respective ground feature types. Therefore, the technology can identify the mangrove forest of the target area by using the remote sensing image identification technology and generate the mangrove forest distribution map. The existing mangrove forest recognition can be divided into a large scale and a small scale according to the recognition scale. The implementation scheme of the small-scale mangrove forest identification is that based on a SPORT5 image, a classification method of SVM (Support Vector Machine) is adopted to analyze and map the mangrove forest type. The large-scale mangrove forest map is a global mangrove forest map based on Landsat and Sentinel-2 images.
Based on the SPORT5 and the color synthetic image, classification drawing of mangroves in the research area is completed by applying an SVM classification method according to four typical mangrove forest remote sensing interpretation marks established by map feature analysis, and finally, precision verification is performed on drawing results by combining random sampling points. But the SVM classifier is at the pixel level, and the utilization of the spatial features is not complete.
In addition, the highest spatial resolution of the SPORT5 image is 2.5 meters, and the accuracy of classification is not high enough. The SPORT5 image data is not open and is expensive to acquire. Landsat, Sentiniel-2 data are open but resolution is low. And the efficiency and accuracy of mangrove forest identification by adopting a machine learning network are low.
Disclosure of Invention
In view of the above, the present invention provides an image recognition method and system, so as to quickly and accurately recognize a mangrove forest in a global scope by using a middle-high resolution remote sensing image classification technology, a deep learning technology and an image segmentation technology, and further obtain a mangrove forest distribution map in the global scope.
In a first aspect, an embodiment of the present invention provides an image recognition method, where the method includes: acquiring RGB remote sensing image data and Sentinel-2 data of a mangrove forest growing region in a global range; the resolution ratio of the RGB remote sensing image data is less than 1 meter; the Sentinel-2 data is data collected by a multispectral imager with the satellite carrying resolution being more than 1 meter, and the resolution is 10 meters; performing up-sampling processing on the Sentinel-2 data to obtain first image data; the resolution of the first image data is the same as that of the RGB remote sensing image data; calculating normalized water body index information and mangrove forest index information of the first image data through band operation, and superposing and fusing the normalized water body index information, the mangrove forest index information, near infrared band information and short wave infrared band information of the first image data and red, green and blue visible band information of the RGB remote sensing image data to obtain second image data; segmenting the RGB remote sensing image data according to preset segmentation parameters to obtain a plurality of segmentation image areas of the RGB remote sensing image data; the types of objects contained in the same segmentation image area are the same, and the types of objects contained in different segmentation image areas are different; the preset segmentation parameters comprise segmentation scale, shape factor and smoothness; removing segmented image areas which do not contain mangrove forest objects in the plurality of segmented image areas of the RGB remote sensing image data according to the Sentinel-2 data, and then carrying out binarization processing to obtain label image data; tiling the RGB remote sensing image data, the second image data and the label image data to obtain an initial sample set containing a plurality of tiles with the same size; and according to an image recognition model which is trained by the initial sample set in advance, predicting the mangrove forest object of the remote sensing image data to be predicted to obtain a mangrove forest prediction result of the remote sensing image data to be predicted.
As a possible implementation, the removing, according to the Sentinel-2 data, a segmented image region not including a mangrove object from among a plurality of segmented image regions of the RGB remote sensing image data, and performing binarization processing to obtain label image data includes: manually labeling a category label for each segmented image area according to the Sentinel-2 data and the type of the object contained in each segmented image area to obtain first label image data; the types of the objects are divided into mangrove forests and non-mangrove forests, the mangroves correspond to the first class labels, and the non-mangroves correspond to the second class labels; removing a segmented image area with a first class label in the first label image data to obtain second label image data; and carrying out binarization processing on the second label image data to obtain the label image data.
As a possible implementation, the acquiring RGB remote sensing image data and Sentinel-2 data of the mangrove forest growing region in the global scope includes: acquiring RGB remote sensing image data of a global mangrove forest growing area; and acquiring the Sentinel-2 data of the global mangrove forest growing region.
As one possible implementation, the training of the image recognition model includes: dividing the initial sample set into a training set, a verification set and a test set according to a preset proportional relation; setting model training parameters; the model training parameters comprise training batch, learning rate and iteration times; performing iterative training on the semantic segmentation model by using the training set, and verifying the precision of the semantic segmentation model after each iterative training by using the verification set; the precision verification indexes of the training set and the verification set about the semantic segmentation model are a first preset loss function and a first intersection ratio function; stopping training until the precision verification indexes of the training set and the verification set about the semantic segmentation model are stable and meet a model convergence condition, and obtaining the image recognition model; testing the trained semantic segmentation model by using the test set; and the test indexes of the training set and the test set about the semantic segmentation model are a second preset loss function and a second intersection ratio function.
As a possible implementation, before the upsampling processing is performed on the Sentinel-2 data to obtain the first image data, the method further includes: and carrying out cloud removing processing on the RGB remote sensing image data and the Sentinel-2 data.
In a second aspect, an embodiment of the present invention further provides an image recognition system, where the system includes: the data acquisition module is used for acquiring RGB remote sensing image data and Sentinel-2 data of a mangrove forest growing region in a global range; the resolution ratio of the RGB remote sensing image data is less than 1 meter; the Sentinel-2 data is data collected by a multispectral imager with the satellite carrying resolution being more than 1 meter, and the resolution is 10 meters; the data preprocessing module is used for performing up-sampling processing on the Sentinel-2 data to obtain first image data; the resolution of the first image data is the same as that of the RGB remote sensing image data; the information processing module is used for calculating the normalized water body index information and the mangrove forest index information of the first image data through band operation, and superposing and fusing the normalized water body index information, the mangrove forest index information, the near-infrared band information and the short-wave infrared band information of the first image data with the red-green-blue three visible light band information of the RGB remote sensing image data to obtain second image data; the data segmentation module is used for segmenting the RGB remote sensing image data according to preset segmentation parameters to obtain a plurality of segmentation image areas of the RGB remote sensing image data; the types of objects contained in the same segmentation image area are the same, and the types of objects contained in different segmentation image areas are different; the preset segmentation parameters comprise segmentation scale, shape factor and smoothness; the label data making module is used for removing segmented image areas which do not contain mangrove objects in the multiple segmented image areas of the RGB remote sensing image data according to the Sentinel-2 data, and then carrying out binarization processing to obtain label image data; the initial sample set forming module is used for tiling the RGB remote sensing image data, the second image data and the label image data to obtain an initial sample set containing a plurality of tiles with the same size; and the model prediction module is used for predicting the mangrove forest object of the remote sensing image data to be predicted according to the image recognition model which is trained by the initial sample set in advance to obtain the mangrove forest prediction result of the remote sensing image data to be predicted.
As a possible implementation, the tag data making module is further configured to: manually labeling a category label for each segmented image area according to the Sentinel-2 data and the type of the object contained in each segmented image area to obtain first label image data; the types of the objects are divided into mangrove forests and non-mangrove forests, the mangroves correspond to the first class labels, and the non-mangroves correspond to the second class labels; removing a segmented image area with a first class label in the first label image data to obtain second label image data; and carrying out binarization processing on the second label image data to obtain the label image data.
As a possible implementation, the data obtaining module is further configured to: acquiring RGB remote sensing image data of a global mangrove forest growing area; and acquiring the Sentinel-2 data of the global mangrove forest growing region.
As a possible implementation, the system further comprises: the model training module is used for dividing the initial sample set into a training set, a verification set and a test set according to a preset proportional relation; setting model training parameters; the model training parameters comprise training batch, learning rate and iteration times; performing iterative training on the semantic segmentation model by using the training set, and verifying the precision of the semantic segmentation model after each iterative training by using the verification set; the precision verification indexes of the training set and the verification set about the semantic segmentation model are a first preset loss function and a first intersection ratio function; until the precision verification indexes of the training set and the verification set about the semantic segmentation model are stable, testing the trained semantic segmentation model by using the test set; the test indexes of the training set and the test set about the semantic segmentation model are a second preset loss function and a second intersection ratio function; and stopping training until the test indexes of the training set and the test set about the semantic segmentation model meet a model convergence condition to obtain the image recognition model.
As a possible implementation, the data preprocessing module is further configured to: and before the Sentinel-2 data is subjected to up-sampling processing to obtain first image data, carrying out cloud removing processing on the RGB remote sensing image data and the Sentinel-2 data.
The embodiment of the invention provides an image identification method and a system, RGB remote sensing image data and Sentinel-2 data of a mangrove forest growing area in a global range are firstly obtained, the Sentinel-2 data is subjected to up-sampling processing to obtain first image data, normalized water index information and mangrove forest index information of the first image data are calculated through band operation, the normalized water index information, the mangrove forest index information, near infrared band information and short wave infrared band information of the first image data are superposed and fused with red, green and blue three visible light band information of the RGB remote sensing image data to obtain second image data, then the RGB remote sensing image data are segmented according to preset segmentation parameters to obtain a plurality of segmentation image areas of the RGB remote sensing image data, the segmentation image areas of the RGB remote sensing image data which do not contain mangrove forest objects are removed according to the Sentinel-2 data, and finally, predicting the mangrove forest object of the remote sensing image data to be predicted according to an image recognition model which is trained by the initial sample set in advance, thereby obtaining a mangrove forest prediction result of the remote sensing image data to be predicted. By adopting the technology, the sub-meter RGB remote sensing image data and the non-sub-meter Sentinel-2 data are used as initial image data, so that the high classification precision can be ensured; when label image data are manufactured, the image is segmented in a semi-automatic mode, so that the time for manufacturing a sample is shortened; the deep learning model is adopted to identify the mangrove forest, so that the incompleteness caused by artificial identification of the mangrove forest is reduced, and meanwhile, the accuracy and efficiency of mangrove forest identification are greatly improved, thereby providing technical support for large-scale and large-scale fine mapping.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
Fig. 1 is a schematic flowchart of an image recognition method according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart illustrating another image recognition method according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating examples of loss dynamics in an embodiment of the present invention;
FIG. 4 is an exemplary diagram of a dynamic MIoU change in an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an image recognition system according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of another image recognition system according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the embodiments, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Currently, existing mangrove forest identification can be divided into a large scale and a small scale according to the identification scale. The implementation scheme of the small-scale mangrove forest identification is that based on a SPORT5 image, a classification method of SVM (Support Vector Machine) is adopted to analyze and map the mangrove forest type. The large-scale mangrove forest map is a global mangrove forest map based on Landsat and Sentinel-2 images. Based on the SPORT5 and the color synthetic image, classification drawing of mangroves in the research area is completed by applying an SVM classification method according to four typical mangrove forest remote sensing interpretation marks established by map feature analysis, and finally, precision verification is performed on drawing results by combining random sampling points. But the SVM classifier is at the pixel level, and the utilization of the spatial features is not complete. In addition, the highest spatial resolution of the SPORT5 image is 2.5 meters, and the accuracy of classification is not high enough. The SPORT5 image data is not open and is expensive to acquire. Landsat, Sentiniel-2 data are open but resolution is low. And the efficiency and accuracy of mangrove forest identification by adopting a machine learning network are low.
Based on the method and the system for identifying the image, which are provided by the embodiment of the invention, the mangrove forest in the global scope can be quickly and accurately identified by utilizing a medium-high resolution remote sensing image classification technology, a deep learning technology and an image segmentation technology, so that a mangrove forest distribution map in the global scope can be obtained.
To facilitate understanding of the embodiment, first, an image recognition method disclosed in the embodiment of the present invention is described in detail, referring to a flowchart of the image recognition method shown in fig. 1, where the method may include the following steps:
s102, acquiring RGB remote sensing image data and Sentinel-2 data of a mangrove forest growing area in a global range; the resolution ratio of the RGB remote sensing image data is less than 1 meter; the Sentinel-2 data is data collected by a multispectral imager with the satellite carrying resolution being more than 1 meter, and the resolution is 10 meters.
The RGB remote sensing image data and the Sentinel-2 data may be obtained by crawling over a network, or obtained from a resource stored locally in advance, and may be determined by itself according to actual needs, which is not limited. For example, the RGB remote sensing image data is from Google Earth.
Step S104, performing up-sampling processing on the Sentinel-2 data to obtain first image data; and the resolution of the first image data is the same as that of the RGB remote sensing image data.
Because the resolution ratio of RGB remote sensing image data is less than 1 meter (resolution ratio is the sub-meter level), and the resolution ratio of Sentinel-2 data is greater than 1 meter (resolution ratio is the non-sub-meter level), in order to promote non-sub-meter level resolution ratio to sub-meter level resolution ratio, can carry out the upsampling to Sentinel-2 data, obtain the first image data of sub-meter level resolution ratio to guarantee that the resolution ratio of the image data that obtains in the follow-up image data processing procedure is the sub-meter level.
And S106, calculating the normalized water body index information and the mangrove forest index information of the first image data through band operation, and superposing and fusing the normalized water body index information, the mangrove forest index information, the near-infrared band information and the short-wave infrared band information of the first image data with the red-green-blue visible band information of the RGB remote sensing image data to obtain second image data.
The normalized water body index information is usually MNDWI and is used for distinguishing water bodies from non-water bodies; the mangrove forest index information is usually WFI and the like, and is used for distinguishing mangroves from non-mangroves.
Step S108, segmenting the RGB remote sensing image data according to preset segmentation parameters to obtain a plurality of segmentation image areas of the RGB remote sensing image data; the types of objects contained in the same segmentation image area are the same, and the types of objects contained in different segmentation image areas are different; the preset segmentation parameters include segmentation scale, shape factor and smoothness.
Generally, the higher the resolution, the smaller the division scale, and thus the setting value of the division scale needs to be determined according to the resolution of the RGB remote sensing image data. In addition, the setting values of the shape factor and the smoothness may be determined by themselves according to actual needs, and are not limited thereto.
After the RGB remote sensing image data is segmented, objects included in each segmented image region may be vegetation, rivers, soil, or the like, the types of objects included in the same segmented image region are the same, and the types of objects included in different segmented image regions are different.
And step S110, according to the Sentinel-2 data, removing the segmented image areas which do not contain the mangrove forest object from the plurality of segmented image areas of the RGB remote sensing image data, and then performing binarization processing to obtain label image data.
Specifically, since the Sentinel-2 data includes a mangrove forest object, the label image data can be obtained by referring to a mangrove forest growing region in the Sentinel-2 data, manually removing a divided image region not including the mangrove forest object from the divided RGB remote sensing image data, and then performing binarization processing.
And S112, tiling the RGB remote sensing image data, the second image data and the label image data to obtain an initial sample set containing a plurality of tiles with the same size.
In order to facilitate the training of the subsequent deep learning model, the RGB remote sensing image data, the second image data and the label image data may be cut into a plurality of tiles with equal sizes, and then each tile is used as an image sample to form the initial sample set.
And S114, predicting the mangrove forest object of the remote sensing image data to be predicted according to the image recognition model which is trained by the initial sample set in advance to obtain the mangrove forest prediction result of the remote sensing image data to be predicted.
The image recognition model can be obtained by training an initial deep learning model, and the specific architecture of the initial deep learning model can be determined according to actual needs without limitation.
In order to improve the accuracy of the prediction of the image recognition model, a spatial attention mechanism can be introduced into the image recognition model, more weight is allocated to the mangrove forest features and less weight is allocated to the non-mangrove forest features in the classification process, so that the accuracy of mangrove forest recognition is improved. For example, a spatial attention module is designed on one or more layers of the initial semantic segmentation model, and the spatial attention module is used for assigning corresponding weights to the mangrove forest features and the non-mangrove forest features by adopting a spatial attention mechanism.
The embodiment of the invention provides an image identification method, which comprises the steps of firstly obtaining RGB remote sensing image data and Sentinel-2 data of a mangrove forest growing area in the global range, carrying out up-sampling processing on the Sentinel-2 data to obtain first image data, calculating normalized water index information and mangrove forest index information of the first image data through band operation, carrying out superposition fusion on the normalized water index information, the mangrove forest index information, near infrared band information and short wave infrared band information of the first image data and red, green and blue visible light band information of the RGB remote sensing image data to obtain second image data, then segmenting the RGB remote sensing image data according to preset segmentation parameters to obtain a plurality of segmentation image areas of the RGB remote sensing image data, removing the segmentation image areas which do not contain mangrove forest objects in the plurality of segmentation image areas of the RGB remote sensing image data according to the Sentinel-2 data, and finally, predicting the mangrove forest object of the remote sensing image data to be predicted according to an image recognition model which is trained by the initial sample set in advance, thereby obtaining a mangrove forest prediction result of the remote sensing image data to be predicted. By adopting the technology, the sub-meter RGB remote sensing image data and the non-sub-meter Sentinel-2 data are used as initial image data, so that the high classification precision can be ensured; when label image data are manufactured, the image is segmented in a semi-automatic mode, so that the time for manufacturing a sample is shortened; the deep learning model is adopted to identify the mangrove forest, so that the incompleteness caused by artificial identification of the mangrove forest is reduced, and meanwhile, the accuracy and efficiency of mangrove forest identification are greatly improved, thereby providing technical support for large-scale and large-scale fine mapping.
In addition to the image recognition method, for convenience of operation, the step S110 (i.e. removing, according to Sentinel-2 data, a segmented image region not including a mangrove forest object from the plurality of segmented image regions of the RGB remote sensing image data, and then performing binarization processing to obtain label image data) may include:
(11) manually labeling a category label for each segmented image area according to the Sentinel-2 data and the type of an object contained in each segmented image area to obtain first label image data; the types of the objects are divided into mangrove forests and non-mangrove forests, the mangroves correspond to the first class labels, and the non-mangroves correspond to the second class labels.
(12) And removing the segmented image area with the first class label in the first label image data to obtain second label image data.
(13) And carrying out binarization processing on the second label image data to obtain label image data.
For example, for a plurality of segmented image regions of RGB remote sensing image data, a first class label 2 is labeled for a segmented image region containing a mangrove forest, a second class label 1 is labeled for a segmented image region containing only a non-mangrove forest, then the segmented image region with the second class label 1 is removed, and the remaining segmented image region with the first class label 2 is subjected to binarization processing, thereby obtaining label image data.
On the basis of the image identification method, for convenience of operation, the step S102 (i.e. acquiring RGB remote sensing image data and Sentinel-2 data of the world-wide mangrove forest growing region) may include:
(21) and acquiring RGB remote sensing image data of a global mangrove forest growing area.
(22) And acquiring the Sentinel-2 data of the global mangrove forest growing region.
For example, RGB remote sensing image data with the resolution of less than 1 meter is downloaded in a water warp notes, and Sentinel-2 data with the resolution of 10 meters is downloaded in a remote sensing cloud computing platform GEE (Google Earth Engine).
As a possible implementation, the training of the image recognition model may include:
(31) and dividing the initial sample set into a training set, a verification set and a test set according to a preset proportional relation.
The preset proportional relationship may be determined according to actual requirements, for example, the proportional relationship among the training set, the verification set, and the test set is 6:2:2, 7:2:1, 8:1:1, and the like, which is not limited.
(32) Setting model training parameters; the model training parameters comprise training batch, learning rate and iteration times.
The training batch, the learning rate, and the number of iterations may be determined according to actual needs, for example, the training batch (i.e., batch _ size) is set to 15, 20, or 30, etc., the learning rate (i.e., learning) is set to 0.001, etc., and the number of iterations (i.e., epoch) is set to 120, 150, or 200, etc., which is not limited herein.
(33) Performing iterative training on the semantic segmentation model by using a training set, and verifying the precision of the semantic segmentation model after each iterative training by using a verification set; the precision verification indexes of the training set and the verification set relative to the semantic segmentation model are a first preset loss function and a first intersection ratio function.
The semantic segmentation model can adopt deplab v1, deplab v2, deplab v3, deplab v3+ and the like, and can be determined by self according to actual needs, and is not limited.
The first preset loss function and the first cross-over ratio function may be determined according to actual needs, for example, the first preset loss function adopts cross entropy loss, softmax loss, or the like, and the first cross-over ratio function adopts MIoU, or the like, which is not limited herein.
(34) Stopping training until the precision verification indexes of the training set and the verification set about the semantic segmentation model are stable and meet the model convergence condition to obtain an image recognition model; testing the trained semantic segmentation model by using a test set; and the test indexes of the training set and the test set about the semantic segmentation model are a second preset loss function and a second intersection ratio function.
The second predetermined loss function and the second intersection ratio function are similar to the first predetermined loss function and the first intersection ratio function, and are not described again.
The above-mentioned model convergence condition may include at least one of: the first preset loss function is smaller than a preset smaller value or stabilizes around a lower value, the first intersection ratio function is larger than a preset value close to 1 or stabilizes around a value close to 1, and the iteration number exceeds a set maximum iteration number.
On the basis of the image recognition method, in order to further enhance the effectiveness of the image data, before performing the step S104 (i.e., performing upsampling processing on Sentinel-2 data to obtain first image data), the method may further include: and carrying out cloud removing treatment on the RGB remote sensing image data and the Sentinel-2 data.
Based on the above image recognition method, the embodiment of the present invention further discloses another image recognition method, as shown in fig. 2, the method may include the following steps:
step S202, RGB remote sensing image data with the resolution of 0.56 m are downloaded in the water menstruation, and Sentinel-2 data with the resolution of 10 m are downloaded in GEE.
And step S204, carrying out cloud removing processing on the RGB remote sensing image data and the Sentinel-2 data.
And step S206, performing up-sampling processing on the Sentinel-2 data by using a resampling tool of the ARCGIS software to obtain first image data with the resolution of 0.56 m.
And S208, calculating MNDWI index information and WFI index information of the first image data through band operation, and superposing and fusing the MNDWI index information, the WFI index information, the near-infrared band information B8 and the short-wave infrared band information B11 of the first image data and the red-green-blue three visible light band information of the RGB remote sensing image data to obtain second image data.
Table 1 shows near-infrared band information B8, short-wave infrared band information B11, and short-wave infrared band information B12, respectively, and table 2 shows the calculation manners of MNDWI index information and WFI index information, respectively.
TABLE 1 band information
TABLE 2 index information
Step S210, setting segmentation parameters through eCoginization software, and performing multi-scale optimized segmentation on RGB remote sensing image data according to the set segmentation parameters to obtain a plurality of segmentation image areas of the RGB remote sensing image data; the types of objects contained in the same segmentation image area are the same, and the types of objects contained in different segmentation image areas are different; the preset segmentation parameters include segmentation scale, shape factor and smoothness.
The segmentation scale can be set to a value between 30 and 100, such as 50; the shape factor can be set to a value between 0 and 1, such as setting the shape factor to 0.4; the smoothness may be set to a value between 0 and 1, such as 0.5. In addition, layer names, band weights, and the like can also be set through the eCogination software.
And S212, overlapping the segmented RGB remote sensing image data with Sentinel-2 data, removing segmented image areas of the segmented RGB remote sensing image data, which do not contain mangrove forest objects, and performing binarization processing to obtain label image data.
Step S214, dividing each of the RGB remote sensing image data, the second image data, and the label image data into a plurality of 256 × 256 (or 512 × 512) tiles, and obtaining an initial sample set including the plurality of tiles.
Step S216, extracting 60% of samples which are uniformly distributed in space from the initial sample set to form a training set, and extracting samples from the rest 40% of samples in the initial sample set to respectively form a verification set and a test set; wherein, the proportion relation between the training set and the verification set is 4: 1.
Step S218, setting the training batch to be 20, setting the learning rate to be 0.001 and setting the iteration number to be 150 (or 200); iteratively training the Deeplab v3+ model by using a training set, and verifying the precision of the Deeplab v3+ model after each iterative training by using a verification set; the precision verification indexes of the training set and the verification set on the Deeplab v3+ model are cross entropy loss and MIoU; stopping training until the precision verification indexes of the training set and the verification set about the Deeplab v3+ model are stable and meet the convergence condition of the model to obtain a mangrove forest recognition model; testing the trained Deeplab v3+ model by using a test set; wherein, the test indexes of the training set and the test set about the Deeplab v3+ model are cross entropy loss and MIoU.
In the training process, the change of the cross entropy loss and the MIoU can be recorded, and a corresponding image is generated. For example, fig. 3, the horizontal axis represents the number of iterations and the vertical axis represents the cross entropy loss; for example, in fig. 4, the horizontal axis represents the number of iterations and the vertical axis represents MIoU.
The above-mentioned model convergence condition is similar to the above-mentioned related contents, and is not described herein again.
And S220, predicting the mangrove forest object of the remote sensing image data to be predicted by using the trained mangrove forest recognition model to obtain a mangrove forest prediction result of the remote sensing image data to be predicted.
Specifically, after a trained mangrove forest identification model is obtained, the remote sensing image data to be predicted is cut into a plurality of 256 × 256 (or 512 × 512) tiles, then each tile is respectively input into the mangrove forest identification model, the predicted image data of 256 × 256 (or 512 × 512) corresponding to each tile is output through the mangrove forest identification model, then the predicted image data corresponding to all the tiles are spliced and coordinated, and the mangrove forest predicted image data of the remote sensing image data to be predicted is obtained.
The mangrove forest prediction result of the remote sensing image data to be predicted can be verified by visual interpretation, and therefore the prediction precision of the mangrove forest recognition model is calculated.
Based on the image recognition method, an embodiment of the present invention further provides an image recognition system, as shown in fig. 5, the system includes the following modules:
the data acquisition module 502 is used for acquiring RGB remote sensing image data and Sentinel-2 data of a mangrove forest growing region in a global range; the resolution ratio of the RGB remote sensing image data is less than 1 meter; the Sentinel-2 data is data collected by a multispectral imager with the satellite carrying resolution being more than 1 meter, and the resolution is 10 meters.
A data preprocessing module 504, configured to perform upsampling processing on the Sentinel-2 data to obtain first image data; and the resolution of the first image data is the same as that of the RGB remote sensing image data.
The information processing module 506 is configured to calculate normalized water body index information and mangrove forest index information of the first image data through band calculation, and superimpose and fuse the normalized water body index information, the mangrove forest index information, near-infrared band information, and short-wave infrared band information of the first image data and red, green, and blue visible band information of the RGB remote sensing image data to obtain second image data.
The data segmentation module 508 is configured to segment the RGB remote sensing image data according to preset segmentation parameters to obtain a plurality of segmented image regions of the RGB remote sensing image data; the types of objects contained in the same segmentation image area are the same, and the types of objects contained in different segmentation image areas are different; the preset segmentation parameters comprise segmentation scale, shape factor and smoothness.
And a label data making module 510, configured to remove, according to the Sentinel-2 data, a segmented image region that does not include a mangrove forest object from the multiple segmented image regions of the RGB remote sensing image data, and perform binarization processing to obtain label image data.
An initial sample set forming module 512, configured to perform tiling on the RGB remote sensing image data, the second image data, and the label image data to obtain an initial sample set including a plurality of tiles with equal sizes.
And the model prediction module 514 is configured to perform prediction of a mangrove forest object on the remote sensing image data to be predicted according to the image recognition model trained by using the initial sample set in advance, so as to obtain a mangrove forest prediction result of the remote sensing image data to be predicted.
The embodiment of the invention provides an image identification system, which comprises the steps of firstly obtaining RGB remote sensing image data and Sentinel-2 data of a mangrove forest growing area in the global range, carrying out up-sampling treatment on the Sentinel-2 data to obtain first image data, calculating normalized water index information and mangrove forest index information of the first image data through band operation, carrying out superposition fusion on the normalized water index information, the mangrove forest index information, near infrared band information and short wave infrared band information of the first image data and red, green and blue three visible light band information of the RGB remote sensing image data to obtain second image data, then segmenting the RGB remote sensing image data according to preset segmentation parameters to obtain a plurality of segmentation image areas of the RGB remote sensing image data, removing the segmentation image areas which do not contain mangrove forest objects in the plurality of segmentation image areas of the RGB remote sensing image data according to the Sentinel-2 data, and finally, predicting the mangrove forest object of the remote sensing image data to be predicted according to an image recognition model which is trained by the initial sample set in advance, thereby obtaining a mangrove forest prediction result of the remote sensing image data to be predicted. By adopting the technology, the sub-meter RGB remote sensing image data and the non-sub-meter Sentinel-2 data are used as initial image data, so that the high classification precision can be ensured; when label image data are manufactured, the image is segmented in a semi-automatic mode, so that the time for manufacturing a sample is shortened; the deep learning model is adopted to identify the mangrove forest, so that the incompleteness caused by artificial identification of the mangrove forest is reduced, and meanwhile, the accuracy and efficiency of mangrove forest identification are greatly improved, thereby providing technical support for large-scale and large-scale fine mapping.
The tag data creation module 510 is further configured to: manually labeling a category label for each segmented image area according to the Sentinel-2 data and the type of the object contained in each segmented image area to obtain first label image data; the types of the objects are divided into mangrove forests and non-mangrove forests, the mangroves correspond to the first class labels, and the non-mangroves correspond to the second class labels; removing a segmented image area with a first class label in the first label image data to obtain second label image data; and carrying out binarization processing on the second label image data to obtain the label image data.
The data obtaining module 502 is further configured to: acquiring RGB remote sensing image data of a global mangrove forest growing area; and acquiring the Sentinel-2 data of the global mangrove forest growing region.
The data preprocessing module 504 is further configured to: and before the Sentinel-2 data is subjected to up-sampling processing to obtain first image data, carrying out cloud removing processing on the RGB remote sensing image data and the Sentinel-2 data.
Based on the image recognition system, another image recognition system is further provided in the embodiments of the present invention, as shown in fig. 6, the system further includes:
a model training module 516, configured to divide the initial sample set into a training set, a verification set, and a test set according to a preset proportional relationship; setting model training parameters; the model training parameters comprise training batch, learning rate and iteration times; performing iterative training on the semantic segmentation model by using the training set, and verifying the precision of the semantic segmentation model after each iterative training by using the verification set; the precision verification indexes of the training set and the verification set about the semantic segmentation model are a first preset loss function and a first intersection ratio function; stopping training until the precision verification indexes of the training set and the verification set about the semantic segmentation model are stable and meet a model convergence condition, and obtaining the image recognition model; testing the trained semantic segmentation model by using the test set; and the test indexes of the training set and the test set about the semantic segmentation model are a second preset loss function and a second intersection ratio function.
The image recognition system provided by the embodiment of the present invention has the same implementation principle and technical effect as the foregoing method embodiments, and for brief description, no mention is made in the system embodiments, and reference may be made to the corresponding contents in the foregoing image recognition method embodiments.
Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present invention, which are used for illustrating the technical solutions of the present invention and not for limiting the same, and the protection scope of the present invention is not limited thereto, although the present invention is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the embodiments of the present invention, and they should be construed as being included therein. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. An image recognition method, characterized in that the method comprises:
acquiring RGB remote sensing image data and Sentinel-2 data of a mangrove forest growing region in a global range; the resolution ratio of the RGB remote sensing image data is less than 1 meter; the Sentinel-2 data is data collected by a multispectral imager with the resolution of more than 1 meter carried by a Sentinel-2 satellite, and the resolution is 10 meters;
performing up-sampling processing on the Sentinel-2 data to obtain first image data; the resolution of the first image data is the same as that of the RGB remote sensing image data;
calculating normalized water body index information and mangrove forest index information of the first image data through band operation, and superposing and fusing the normalized water body index information, the mangrove forest index information, near infrared band information and short wave infrared band information of the first image data and red, green and blue visible band information of the RGB remote sensing image data to obtain second image data;
segmenting the RGB remote sensing image data according to preset segmentation parameters to obtain a plurality of segmentation image areas of the RGB remote sensing image data; the types of objects contained in the same segmentation image area are the same, and the types of objects contained in different segmentation image areas are different; the preset segmentation parameters comprise segmentation scale, shape factor and smoothness;
removing segmented image areas which do not contain mangrove forest objects in the plurality of segmented image areas of the RGB remote sensing image data according to the Sentinel-2 data, and then carrying out binarization processing to obtain label image data;
tiling the RGB remote sensing image data, the second image data and the label image data to obtain an initial sample set containing a plurality of tiles with the same size;
and according to an image recognition model which is trained by the initial sample set in advance, predicting the mangrove forest object of the remote sensing image data to be predicted to obtain a mangrove forest prediction result of the remote sensing image data to be predicted.
2. The image recognition method according to claim 1, wherein the obtaining label image data by removing, according to the Sentinel-2 data, a segmented image region that does not include a mangrove forest object from among the plurality of segmented image regions of the RGB remote-sensing image data and performing binarization processing includes:
manually labeling a category label for each segmented image area according to the Sentinel-2 data and the type of the object contained in each segmented image area to obtain first label image data; the types of the objects are divided into mangrove forests and non-mangrove forests, the mangroves correspond to the first class labels, and the non-mangroves correspond to the second class labels;
removing a segmented image area with a first class label in the first label image data to obtain second label image data;
and carrying out binarization processing on the second label image data to obtain the label image data.
3. The image recognition method according to claim 1, wherein the acquiring of RGB remote sensing image data and Sentinel-2 data of the globally-wide mangrove forest growing region comprises:
acquiring RGB remote sensing image data of a mangrove forest growing region in a global range;
and acquiring the Sentinel-2 data of the mangrove forest growing region in the global range.
4. The image recognition method of claim 1, wherein the training of the image recognition model comprises:
dividing the initial sample set into a training set, a verification set and a test set according to a preset proportional relation;
setting model training parameters; the model training parameters comprise training batch, learning rate and iteration times;
performing iterative training on the semantic segmentation model by using the training set, and verifying the precision of the semantic segmentation model after each iterative training by using the verification set; the precision verification indexes of the training set and the verification set about the semantic segmentation model are a first preset loss function and a first intersection ratio function;
stopping training until the precision verification indexes of the training set and the verification set about the semantic segmentation model are stable and meet a model convergence condition, and obtaining the image recognition model; testing the trained semantic segmentation model by using the test set; and the test indexes of the training set and the test set about the semantic segmentation model are a second preset loss function and a second intersection ratio function.
5. The image recognition method according to claim 1, wherein before the upsampling process is performed on the Sentinel-2 data to obtain first image data, the method further comprises:
and carrying out cloud removing processing on the RGB remote sensing image data and the Sentinel-2 data.
6. An image recognition system, the system comprising:
the data acquisition module is used for acquiring RGB remote sensing image data and Sentinel-2 data of a mangrove forest growing region in a global range; the resolution ratio of the RGB remote sensing image data is less than 1 meter; the Sentinel-2 data is data collected by a multispectral imager with the satellite carrying resolution being more than 1 meter, and the resolution is 10 meters;
the data preprocessing module is used for performing up-sampling processing on the Sentinel-2 data to obtain first image data; the resolution of the first image data is the same as that of the RGB remote sensing image data;
the information processing module is used for calculating the normalized water body index information and the mangrove forest index information of the first image data through band operation, and superposing and fusing the normalized water body index information, the mangrove forest index information, the near-infrared band information and the short-wave infrared band information of the first image data with the red-green-blue three visible light band information of the RGB remote sensing image data to obtain second image data;
the data segmentation module is used for segmenting the RGB remote sensing image data according to preset segmentation parameters to obtain a plurality of segmentation image areas of the RGB remote sensing image data; the types of objects contained in the same segmentation image area are the same, and the types of objects contained in different segmentation image areas are different; the preset segmentation parameters comprise segmentation scale, shape factor and smoothness;
the label data making module is used for removing segmented image areas which do not contain mangrove objects in the multiple segmented image areas of the RGB remote sensing image data according to the Sentinel-2 data, and then carrying out binarization processing to obtain label image data;
the initial sample set forming module is used for tiling the RGB remote sensing image data, the second image data and the label image data to obtain an initial sample set containing a plurality of tiles with the same size;
and the model prediction module is used for predicting the mangrove forest object of the remote sensing image data to be predicted according to the image recognition model which is trained by the initial sample set in advance to obtain the mangrove forest prediction result of the remote sensing image data to be predicted.
7. The image recognition system of claim 6, wherein the tag data production module is further configured to:
manually labeling a category label for each segmented image area according to the Sentinel-2 data and the type of the object contained in each segmented image area to obtain first label image data; the types of the objects are divided into mangrove forests and non-mangrove forests, the mangroves correspond to the first class labels, and the non-mangroves correspond to the second class labels;
removing a segmented image area with a first class label in the first label image data to obtain second label image data;
and carrying out binarization processing on the second label image data to obtain the label image data.
8. The image recognition system of claim 6, wherein the data acquisition module is further configured to:
acquiring RGB remote sensing image data of a mangrove forest growing region in a global range;
and acquiring the Sentinel-2 data of the mangrove forest growing region in the global range.
9. The image recognition system of claim 6, further comprising:
the model training module is used for dividing the initial sample set into a training set, a verification set and a test set according to a preset proportional relation; setting model training parameters; the model training parameters comprise training batch, learning rate and iteration times; performing iterative training on the semantic segmentation model by using the training set, and verifying the precision of the semantic segmentation model after each iterative training by using the verification set; the precision verification indexes of the training set and the verification set about the semantic segmentation model are a first preset loss function and a first intersection ratio function; stopping training until the precision verification indexes of the training set and the verification set about the semantic segmentation model are stable and meet a model convergence condition, and obtaining the image recognition model; testing the trained semantic segmentation model by using the test set; and the test indexes of the training set and the test set about the semantic segmentation model are a second preset loss function and a second intersection ratio function.
10. The image recognition system of claim 6, wherein the data pre-processing module is further configured to: and before the Sentinel-2 data is subjected to up-sampling processing to obtain first image data, carrying out cloud removing processing on the RGB remote sensing image data and the Sentinel-2 data.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210623004.XA CN114898097B (en) | 2022-06-01 | 2022-06-01 | Image recognition method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210623004.XA CN114898097B (en) | 2022-06-01 | 2022-06-01 | Image recognition method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114898097A true CN114898097A (en) | 2022-08-12 |
CN114898097B CN114898097B (en) | 2024-05-10 |
Family
ID=82726327
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210623004.XA Active CN114898097B (en) | 2022-06-01 | 2022-06-01 | Image recognition method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114898097B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115601688B (en) * | 2022-12-15 | 2023-02-21 | 中译文娱科技(青岛)有限公司 | Video main content detection method and system based on deep learning |
CN116862836A (en) * | 2023-05-30 | 2023-10-10 | 北京透彻未来科技有限公司 | System and computer equipment for detecting extensive organ lymph node metastasis cancer |
CN117036982A (en) * | 2023-10-07 | 2023-11-10 | 山东省国土空间数据和遥感技术研究院(山东省海域动态监视监测中心) | Method and device for processing optical satellite image of mariculture area, equipment and medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108564587A (en) * | 2018-03-07 | 2018-09-21 | 浙江大学 | A kind of a wide range of remote sensing image semantic segmentation method based on full convolutional neural networks |
CN109543630A (en) * | 2018-11-28 | 2019-03-29 | 苏州中科天启遥感科技有限公司 | Remote sensing image forest land extracting method and system, storage medium, electronic equipment based on deep learning |
US20200026953A1 (en) * | 2018-07-23 | 2020-01-23 | Wuhan University | Method and system of extraction of impervious surface of remote sensing image |
CN110852225A (en) * | 2019-10-31 | 2020-02-28 | 中国地质大学(武汉) | Remote sensing image mangrove forest extraction method and system based on deep convolutional neural network |
WO2021184891A1 (en) * | 2020-03-20 | 2021-09-23 | 中国科学院深圳先进技术研究院 | Remotely-sensed image-based terrain classification method, and system |
-
2022
- 2022-06-01 CN CN202210623004.XA patent/CN114898097B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108564587A (en) * | 2018-03-07 | 2018-09-21 | 浙江大学 | A kind of a wide range of remote sensing image semantic segmentation method based on full convolutional neural networks |
US20200026953A1 (en) * | 2018-07-23 | 2020-01-23 | Wuhan University | Method and system of extraction of impervious surface of remote sensing image |
CN109543630A (en) * | 2018-11-28 | 2019-03-29 | 苏州中科天启遥感科技有限公司 | Remote sensing image forest land extracting method and system, storage medium, electronic equipment based on deep learning |
CN110852225A (en) * | 2019-10-31 | 2020-02-28 | 中国地质大学(武汉) | Remote sensing image mangrove forest extraction method and system based on deep convolutional neural network |
WO2021184891A1 (en) * | 2020-03-20 | 2021-09-23 | 中国科学院深圳先进技术研究院 | Remotely-sensed image-based terrain classification method, and system |
Non-Patent Citations (2)
Title |
---|
李想;刘凯;朱远辉;蒙琳;于晨曦;曹晶晶;: "基于资源三号影像的红树林物种分类研究", 遥感技术与应用, no. 02, 20 April 2018 (2018-04-20) * |
蒙良莉;凌子燕;蒋卫国;钟仕全;陈燕丽;孙明;: "基于Sentinel遥感数据的红树林信息提取研究――以广西茅尾海为例", 地理与地理信息科学, no. 04, 15 July 2020 (2020-07-15) * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115601688B (en) * | 2022-12-15 | 2023-02-21 | 中译文娱科技(青岛)有限公司 | Video main content detection method and system based on deep learning |
CN116862836A (en) * | 2023-05-30 | 2023-10-10 | 北京透彻未来科技有限公司 | System and computer equipment for detecting extensive organ lymph node metastasis cancer |
CN117036982A (en) * | 2023-10-07 | 2023-11-10 | 山东省国土空间数据和遥感技术研究院(山东省海域动态监视监测中心) | Method and device for processing optical satellite image of mariculture area, equipment and medium |
CN117036982B (en) * | 2023-10-07 | 2024-01-09 | 山东省国土空间数据和遥感技术研究院(山东省海域动态监视监测中心) | Method and device for processing optical satellite image of mariculture area, equipment and medium |
Also Published As
Publication number | Publication date |
---|---|
CN114898097B (en) | 2024-05-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Ji et al. | Fully convolutional networks for multisource building extraction from an open aerial and satellite imagery data set | |
US9230168B2 (en) | Automatic generation of built-up layers from high resolution satellite image data | |
CN114898097B (en) | Image recognition method and system | |
CN111738165B (en) | Method for extracting individual plant canopy from high-resolution unmanned aerial vehicle visible light remote sensing image | |
EP3022686B1 (en) | Automatic generation of multi-scale descriptors from overhead imagery through manipulation of alpha-tree data structures | |
CN110084817B (en) | Digital elevation model production method based on deep learning | |
CN111028255B (en) | Farmland area pre-screening method and device based on priori information and deep learning | |
CN114241326B (en) | Progressive intelligent production method and system for ground feature elements of remote sensing images | |
CN114399686A (en) | Remote sensing image ground feature identification and classification method and device based on weak supervised learning | |
CN115223054A (en) | Remote sensing image change detection method based on partition clustering and convolution | |
CN111461006A (en) | Optical remote sensing image tower position detection method based on deep migration learning | |
Atik et al. | Comparative research on different backbone architectures of DeepLabV3+ for building segmentation | |
Soilán et al. | Comparison of heuristic and deep learning-based methods for ground classification from aerial point clouds | |
CN111079807A (en) | Ground object classification method and device | |
CN113298042B (en) | Remote sensing image data processing method and device, storage medium and computer equipment | |
CN116503677A (en) | Wetland classification information extraction method, system, electronic equipment and storage medium | |
Schuegraf et al. | Deep learning for the automatic division of building constructions into sections on remote sensing images | |
Sivagami et al. | Analysis of encoder-decoder based deep learning architectures for semantic segmentation in remote sensing images | |
Norelyaqine et al. | Deep learning for building extraction from high-resolution remote sensing images | |
Höhle et al. | Automated Extraction of Topographic Map Data from Remotely Sensed Imagery by Classification and Cartographic Enhancement: An Introduction to New Mapping Tools | |
Karsli et al. | Automatic building footprint extraction from photogrammetric and LiDAR point clouds using a novel improved-Octree approach | |
CN114549534B (en) | Mining area land utilization identification method, device, equipment and medium | |
CN117115566B (en) | Urban functional area identification method and system by utilizing full-season remote sensing images | |
Papadopoulos et al. | Convolutional Neural Network for Detection of Building Contours Using Multisource Spatial Data | |
Donnelly et al. | Identifying Urban Canopy Coverage from Satellite Imagery Using Convolutional Neural Networks |
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 |