CN117522884B - Ocean remote sensing image semantic segmentation method and device and electronic equipment - Google Patents
Ocean remote sensing image semantic segmentation method and device and electronic equipment Download PDFInfo
- Publication number
- CN117522884B CN117522884B CN202410015646.0A CN202410015646A CN117522884B CN 117522884 B CN117522884 B CN 117522884B CN 202410015646 A CN202410015646 A CN 202410015646A CN 117522884 B CN117522884 B CN 117522884B
- Authority
- CN
- China
- Prior art keywords
- feature
- remote sensing
- scale
- image
- decoding
- 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
- 230000011218 segmentation Effects 0.000 title claims abstract description 67
- 238000000034 method Methods 0.000 title claims abstract description 40
- 230000004927 fusion Effects 0.000 claims abstract description 39
- 238000000605 extraction Methods 0.000 claims abstract description 38
- 238000010586 diagram Methods 0.000 claims description 26
- 238000012545 processing Methods 0.000 claims description 18
- 238000010606 normalization Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 8
- 238000003709 image segmentation Methods 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 7
- 230000008447 perception Effects 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 230000000875 corresponding effect Effects 0.000 description 21
- 230000008569 process Effects 0.000 description 9
- 230000009977 dual effect Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—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
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/52—Scale-space analysis, e.g. wavelet analysis
-
- 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
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Biodiversity & Conservation Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a semantic segmentation method and device for ocean remote sensing images and electronic equipment, wherein the semantic segmentation method comprises the following steps: acquiring a marine remote sensing image to be identified; in a semantic segmentation network of an ocean remote sensing image, global context information is extracted based on crisscross attention to obtain a multi-scale feature map, space detail feature extraction is carried out on the multi-scale feature map to obtain space detail information, cascading multi-scale fusion, feature decoding and double-branch channel attention weighting are carried out on the multi-scale feature map to obtain feature decoding output, and semantic segmentation prediction is carried out after the feature decoding output and the space detail information are spliced to obtain a segmentation result of the ocean remote sensing image. In conclusion, multiscale semantic information is extracted through crisscross attention, redundant features and enhanced detail features are suppressed through double-branch channel attention weighting, shallow space detail information and deep global context information are fused through cascading multiscale fusion, and accurate segmentation of ocean remote sensing image semantics is achieved.
Description
Technical Field
The invention relates to the field of image semantic segmentation, in particular to a semantic segmentation method and device for ocean remote sensing images and electronic equipment.
Background
The ocean remote sensing technology has important significance in aspects of disaster prediction, resource detection, ocean protection and the like, wherein semantic segmentation is important in the field of computer vision as a core technology for analyzing ocean remote sensing images. The semantic segmentation aims at carrying out pixel-by-pixel prediction on an input image to realize region segmentation, and further identifying the semantic information of the category as a segmentation result and overlapping higher-layer semantics. The existing semantic segmentation method is mainly based on FCN, multi-scale, graph convolution and other technologies for feature mining, or uses Transform to cut a remote sensing image into small graph blocks so as to reduce the calculated amount, and captures context information through an attention mechanism to establish the relation between target objects.
However, the existing methods have some problems: first, the problem that the intra-class difference is large and the inter-class difference is small cannot be solved on the convolutional neural network model of the feature coupling. In the marine remote sensing image, for example, the formation process of floating ice and anchor ice is influenced by factors such as temperature, water density and the like, so that the intra-class difference is large and the similarity is low; meanwhile, for example, cargo ships and passenger ships belong to two different semantic categories, but are highly similar in appearance, resulting in small inter-category differences. Although the existing semantic segmentation method can realize pixel-level prediction, global context information cannot be fully mined due to the limitations of receptive fields. Secondly, the target object cannot be accurately described on the feature space with entangled dimensions, the semantic information scale in the marine remote sensing image is seriously entangled, for example, the area of islands in normal conditions is far larger than that of ships, the model can be correctly classified, but if the ships are far larger than that of the islands, the model is difficult to accurately describe the two kinds of semantics. For the feature space scale entanglement condition of the same semantic, the prior art is difficult to accurately identify the multi-scale targets under the same semantic, confusion of understanding of the semantic by a model easily occurs, a plurality of targets with a plurality of scales of the same semantic are regarded as the same object to be modeled, and the target objects cannot be accurately described.
Disclosure of Invention
In view of the foregoing, it is necessary to provide a semantic segmentation method, device and electronic equipment for ocean remote sensing images, which are used for solving the technical problems that the prior art cannot fully mine global context information of the ocean remote sensing images, and is difficult to identify multi-scale targets under the same semantic meaning, so that the accuracy of semantic segmentation of the ocean remote sensing images is low.
In order to solve the above problems, in one aspect, the present invention provides a semantic segmentation method for marine remote sensing images, including:
acquiring a marine remote sensing image to be identified;
Inputting the ocean remote sensing image to be identified into a ocean remote sensing image semantic segmentation network with complete training, performing global context information extraction on the ocean remote sensing image to be identified based on crisscross attention by the ocean remote sensing image semantic segmentation network to obtain a multi-scale feature image, performing space detail feature extraction on the multi-scale feature image to obtain space detail information, performing cascading multi-scale fusion, feature decoding and dual-branch channel attention weighting on the multi-scale feature image to obtain feature decoding output, and performing semantic segmentation prediction after the feature decoding output and the space detail information are spliced to obtain an ocean remote sensing image segmentation result.
Further, extracting global context information of the ocean remote sensing image to be identified based on the crisscross attention to obtain a multi-scale feature map, including:
performing patch embedding on the ocean remote sensing image to be identified to obtain a preliminary feature map;
And carrying out multi-stage crisscross feature extraction on the preliminary feature map step by step to obtain a multi-scale feature map, wherein the multi-scale feature map comprises a first feature map, a second feature map, a third feature map and a fourth feature map.
Further, each stage in the multi-stage cross feature extraction includes a plurality of feature extraction modules including a normalization layer, cross multi-headed attention, jump connections, normalization layer, and multi-layer perception operations.
Further, extracting the space detail features of the multi-scale feature map to obtain space detail information includes:
Taking the first feature map as feature input for extracting first space detail features, taking the second feature map as guide input for extracting the first space detail features, and extracting the space detail features based on residual convolution to obtain first detail features;
Taking the first detail feature as feature input for extracting the second space detail feature, taking the third feature map as guide input for extracting the second space detail feature, and extracting the space detail feature based on residual convolution to obtain the second detail feature;
And taking the second detail feature as a feature input for extracting the third space detail feature, taking the fourth feature map as a guide input for extracting the third space detail feature, and extracting the space detail feature based on residual convolution to obtain the space detail feature.
Further, performing spatial detail features based on residual convolution, including:
performing up-sampling and convolution mapping operation on the guiding input to obtain guiding characteristics;
and splicing the guide features into feature input, and carrying out convolution, normalization, batch regularization and activation functions twice in succession to obtain detail feature output.
Further, performing cascade multi-scale fusion, feature decoding and dual-branch channel attention weighting on the multi-scale feature map to obtain a feature decoding output, including:
performing self-adaptive downsampling and feature fusion on the multi-scale feature map to obtain a fusion feature map corresponding to each decoding stage;
And splicing the fourth feature map step by step with the corresponding fusion feature map, performing feature decoding to obtain feature decoding maps of all levels, and adjusting feature weights of the feature decoding maps based on the attention of the double branch channels to obtain feature decoding output.
Further, performing adaptive downsampling on the multi-scale feature map to obtain a fused feature map corresponding to each decoding stage, including:
Determining the corresponding downsampling times of each multi-scale feature map according to the scale ordering of the multi-scale feature map;
Performing downsampling operation for corresponding times on the multi-scale feature map step by step according to the corresponding downsampling times to obtain multi-stage downsampling features, wherein the multi-stage downsampling features correspond to each decoding stage;
And fusing the multi-level downsampling characteristics corresponding to the same decoding stage, which are obtained by downsampling different multi-scale characteristic diagrams, to obtain a fused characteristic diagram corresponding to each decoding stage.
Further, adjusting feature weights of the feature decoding graph based on the dual-branch channel attention comprises:
Performing preliminary convolution treatment on the feature decoding graph;
Carrying out weight extraction on the feature map subjected to the primary convolution treatment based on SE channel attention to obtain feature weights, and carrying out primary feature weighting on the feature map subjected to the primary convolution treatment according to the feature weights;
And carrying out sparsity calculation on the feature map subjected to the preliminary convolution treatment based on sparse channel attention to obtain channel weights, and carrying out channel feature weighting on the feature map subjected to the preliminary convolution treatment according to the channel weights.
On the other hand, the invention also provides a semantic segmentation device for the marine remote sensing image, which comprises the following steps:
the image acquisition unit is used for acquiring the ocean remote sensing image to be identified;
The semantic segmentation unit is used for inputting the ocean remote sensing image to be identified into a ocean remote sensing image semantic segmentation network with complete training, carrying out global context information extraction on the ocean remote sensing image to be identified by the ocean remote sensing image semantic segmentation network to obtain a multi-scale feature image, carrying out space detail feature extraction on the multi-scale feature image to obtain space detail information, carrying out cascading multi-scale fusion, feature decoding and double-branch channel attention weighting on the multi-scale feature image to obtain feature decoding output, and carrying out semantic segmentation prediction after the feature decoding output and the space detail information are spliced to obtain an ocean remote sensing image segmentation result.
In another aspect, the invention also provides an electronic device comprising a memory and a processor, wherein,
A memory for storing a computer program;
And the processor is coupled with the memory and used for executing the computer program to realize the steps in the semantic segmentation method of the marine remote sensing image.
Compared with the prior art, the beneficial effects of adopting the embodiment are as follows: according to the invention, multiscale semantic information in the marine remote sensing image is extracted through crisscross attention, redundant features and enhanced detail features are suppressed through double-branch channel attention weighting, fine multiscale feature fusion is realized, shallow space detail information and deep global context information are fused through cascade multiscale fusion, and accurate segmentation of the marine remote sensing image semantics is realized.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being evident that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of an embodiment of a semantic segmentation method for marine remote sensing images provided by the invention;
FIG. 2 is an overall network architecture diagram of an embodiment;
FIG. 3 is a cross-attention backbone network block diagram of an embodiment;
FIG. 4 is a network structure diagram of spatial detail feature extraction of an embodiment;
FIG. 5 is a network block diagram of a dual branch channel attention module of an embodiment;
FIG. 6 is a schematic structural diagram of an embodiment of a semantic segmentation device for marine remote sensing images provided by the present invention;
fig. 7 is a schematic structural diagram of an embodiment of an electronic device provided by 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 accompanying drawings in the embodiments of the present invention. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be understood that the drawings of the schematic drawings are not drawn to scale. A flowchart, as used in this disclosure, illustrates operations implemented according to some embodiments of the present invention. It should be appreciated that the operations of the flow diagrams may be implemented out of order and that steps without logical context may be performed in reverse order or concurrently. Moreover, one or more other operations may be added to or removed from the flow diagrams by those skilled in the art under the direction of the present disclosure.
Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor systems and/or microcontroller systems.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the invention. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.
Fig. 1 is a schematic flow chart of an embodiment of a semantic segmentation method for marine remote sensing images, where the semantic segmentation method for marine remote sensing images, as shown in fig. 1, includes:
S101, acquiring a marine remote sensing image to be identified;
S102, inputting the ocean remote sensing image to be identified into a ocean remote sensing image semantic segmentation network with complete training, performing global context information extraction on the ocean remote sensing image to be identified based on crisscross attention by the ocean remote sensing image semantic segmentation network to obtain a multi-scale feature image, performing space detail feature extraction on the multi-scale feature image to obtain space detail information, performing cascading multi-scale fusion, feature decoding and double-branch channel attention weighting on the multi-scale feature image to obtain feature decoding output, and performing semantic segmentation prediction after the feature decoding output and the space detail information are spliced to obtain an ocean remote sensing image segmentation result.
Specifically, in the semantic segmentation method of the marine remote sensing image, multiscale semantic information in the marine remote sensing image is extracted through crisscross attention, redundant features and enhanced detail features are suppressed through double branch channel attention weighting, fine multiscale feature fusion is achieved, shallow space detail information and deep global context information are fused through cascade multiscale fusion, and accurate segmentation of the marine remote sensing image semantic is achieved.
In a specific embodiment of the present invention, global context information extraction is performed on a marine remote sensing image to be identified based on crisscross attention to obtain a multi-scale feature map, including:
performing patch embedding on the ocean remote sensing image to be identified to obtain a preliminary feature map;
And carrying out multi-stage crisscross feature extraction on the preliminary feature map step by step to obtain a multi-scale feature map, wherein the multi-scale feature map comprises a first feature map, a second feature map, a third feature map and a fourth feature map.
In a specific embodiment of the present invention, each stage of the multi-stage cross feature extraction includes a plurality of feature extraction modules including normalization layer, cross-multi-head attention, skip connection, normalization layer, and multi-layer perception operations performed sequentially.
Specifically, fig. 2 is a whole network structure diagram of an embodiment, and as shown in fig. 2, an encoder part of the embodiment is composed of a backbone network based on crisscross attention and a spatial detail feature extraction module. Wherein the backbone network is used for inputting ocean remote sensing images to be identifiedPatch embedding is carried out through Token Embedding operation based on convolution, the size of the patch is reduced, the number of channels of the patch is increased, and a preliminary feature map/> isobtained. Then, FIG. 3 is a cross-attention backbone network structure diagram of an embodiment, as shown in FIG. 3, preliminary feature map/>Four stages of processing are performed, each stage including a plurality CSwin Transformer of feature extraction modules, and each CSwin Transformer module outputs/>, from a previous CSwin Transformer moduleAs input to the module. Input/>, of each CSwin Transformer moduleFirstly, through a normalization and cross multi-head Attention CSwin-Attention, and the obtained characteristics and input/>Establishing a jump connection to obtain an intermediate output/>Subsequently/>The final output/>, of the module is obtained through another normalization layer with jump connection and a multi-layer perception operation MLP layer. Wherein, the crisscross multi-head Attention CSwin-Attention is to input a characteristic diagram/>Through layer normalization processing and then projection to individual/>Head, then divide it into/>And/>Performing self-attention modeling in the horizontal direction and the vertical direction respectively, and re-splicing the obtained outputs to be used as the final output of the module, wherein the final output is specifically shown as follows:
wherein, Is the horizontal direction attention,/>Is the vertical direction attention,/>For projecting the output to the appropriate dimension.
For self-attention modeling in the horizontal direction, the procedure is as follows:
Wherein the first formula represents Uniform division into non-overlapping horizontal blocks/>Their width is/>Each contains/>And feature points. Wherein/>Is the number of horizontal blocks, and/>; For the second formula, the/>, is representedHorizontal attention mechanism of individual head, wherein/>Represents the/>Dimension of individual queries, keys and values, wherein/>，/>And/>Specific self-attention modeling is as follows:
wherein, For inquiry,/>For key,/>For the value of/>Is/>Is a dimension of (c).
In a specific embodiment of the present invention, extracting spatial detail features from a multi-scale feature map to obtain spatial detail information includes:
Taking the first feature map as feature input for extracting first space detail features, taking the second feature map as guide input for extracting the first space detail features, and extracting the space detail features based on residual convolution to obtain first detail features;
Taking the first detail feature as feature input for extracting the second space detail feature, taking the third feature map as guide input for extracting the second space detail feature, and extracting the space detail feature based on residual convolution to obtain the second detail feature;
And taking the second detail feature as a feature input for extracting the third space detail feature, taking the fourth feature map as a guide input for extracting the third space detail feature, and extracting the space detail feature based on residual convolution to obtain the space detail feature.
Specifically, as shown in fig. 2, the spatial detail feature extraction module is composed of three stages, except that the first stage takes the first feature map as input, the other stages take the output of the previous stage as input, and spatial detail information in the shallow high-resolution feature map is extracted for multiple times in a grading manner under the assistance of the corresponding deep semantic information of the lower layer of the backbone network, which is expressed as follows:
wherein, Extracting module for space detail features/>Output of/>For/>Is also/>Output of/>Is backbone network/>And outputting the extracted cross characteristic of each stage.
With a second spatial detail feature extraction moduleFor example, its input is the previous module/>Output/>And a third layer output feature map/>, of the backbone network，/>Responsible for at/>Further based on the output of the (c) in deep semantic information/>Extracting space detail information in the shallow high-resolution characteristic map with the aid of the method.
In a specific embodiment of the present invention, spatial detail features based on residual convolution include:
performing up-sampling and convolution mapping operation on the guiding input to obtain guiding characteristics;
and splicing the guide features into feature input, and carrying out convolution, normalization, batch regularization and activation functions twice in succession to obtain detail feature output.
Specifically, FIG. 4 is a network structure diagram of spatial detail feature extraction of an embodiment, as shown in FIG. 4, toRepresenting low-level feature inputs,/>Representing advanced instructional features,/>The specific flow of (2) is as follows:
wherein, Representing a1 x 1 convolution,/>Representing two consecutive follow/>Batch regularization and activation function/>Is a3 x 3 convolution of (c). The invention will/>, firstUpsampling and oneConvolution maps it to/>. Will subsequently/>Splice to/>And then carrying out two convolution operations, wherein the convolution sum is 3 and the step size is 1, and each convolution operation is followed by normalization operation/>Batch regularization and activation function/>. Wherein the first convolution operation will halve the number of channels of the feature map after aggregation, from/>Reduced to/>While the second convolution operation keeps the number of channels unchanged so that the output can be the same in size and number of channels as the input low-level feature map, can be taken as the next/>And (3) continuously extracting the space detail information with the assistance of the deeper level features output by the next layer of the backbone network.
In a specific embodiment of the present invention, performing cascading multi-scale fusion, feature decoding and dual-branch channel attention weighting on a multi-scale feature map to obtain a feature decoded output, including:
performing self-adaptive downsampling and feature fusion on the multi-scale feature map to obtain a fusion feature map corresponding to each decoding stage;
And splicing the fourth feature map step by step with the corresponding fusion feature map, performing feature decoding to obtain feature decoding maps of all levels, and adjusting feature weights of the feature decoding maps based on the attention of the double branch channels to obtain feature decoding output.
Specifically, the decoder portion of the embodiment obtains a plurality of fusion feature maps by performing adaptive downsampling and feature fusion on the multi-scale feature maps, and the fusion feature maps correspond to each stage of the decoder. Splicing the corresponding fusion feature map to a fourth feature map and each level of feature decoding map input to a decoder in the decoding process, so as to realize fusion of shallow space detail information and deep global context information; and redundant features are suppressed based on double-branch channel attention weighting in the decoding process, so that detail features difficult to extract are enhanced, and fine multi-scale feature fusion is realized.
In a specific embodiment of the present invention, performing adaptive downsampling on a multi-scale feature map to obtain a fused feature map corresponding to each decoding stage includes:
Determining the corresponding downsampling times of each multi-scale feature map according to the scale ordering of the multi-scale feature map;
Performing downsampling operation for corresponding times on the multi-scale feature map step by step according to the corresponding downsampling times to obtain multi-stage downsampling features, wherein the multi-stage downsampling features correspond to each decoding stage;
And fusing the multi-level downsampling characteristics corresponding to the same decoding stage, which are obtained by downsampling different multi-scale characteristic diagrams, to obtain a fused characteristic diagram corresponding to each decoding stage.
Specifically, as shown in fig. 2, the purpose of the cascade multi-scale fusion module is to downsample a large-size low-level feature map, then supplement abundant space detail information into deep semantic features, and the sampling times are determined according to the scale sequence of the multi-scale feature map. For example, a first feature map for a first layer output of a backbone networkFor example, the embodiments are respectively toNo operation, one downsampling, two downsampling and three downsampling are performed, as follows:
wherein, ，/>，/>And/>Pair/>, respectivelyAfter processing, multi-level downsampling characteristics to be aggregated into decoder first, second, third and fourth stages,/>，/>，/>Three downsampling steps are respectively carried out, and the difference is that the number of channels of input and output is different. At the same time, to reduce the number of module parameters, embodiments share parameters for different downsampling processes, such as/>, in one downsampling operationThere are also uses in the secondary downsampling and the tertiary downsampling.
Feature maps for other layers、/>And/>Downsampling process and/>The same difference is that the number of downsampling times decreases in sequence according to the scale of the feature map. After the downsampling process, the embodiment fuses the corresponding multi-level downsampling features obtained from different feature maps in the same decoding stage to obtain a fused feature map corresponding to the decoding stage.
In a specific embodiment of the present invention, adjusting feature weights of a feature decoding graph based on dual branch channel attention includes:
Performing preliminary convolution treatment on the feature decoding graph;
Carrying out weight extraction on the feature map subjected to the primary convolution treatment based on SE channel attention to obtain feature weights, and carrying out primary feature weighting on the feature map subjected to the primary convolution treatment according to the feature weights;
And carrying out sparsity calculation on the feature map subjected to the preliminary convolution treatment based on sparse channel attention to obtain channel weights, and carrying out channel feature weighting on the feature map subjected to the preliminary convolution treatment according to the channel weights.
Specifically, fig. 5 is a network structure diagram of a dual-branch channel attention module of an embodiment, and as shown in fig. 5, in order to implement fine multi-scale feature fusion, the embodiment adjusts feature weights of each feature decoding graph in the decoding process based on the dual-branch channel attention. In dual branch channel attention, one branch is a standard SE channel attention and the other branch is a sparse channel attention based on feature channel sparsity.
In the SE channel attention module, a convolution kernel with a size of 3×3 and a step size of 1 is first used to pair the feature map input to the SE channel attention moduleProcessing is performed and the convolution is followed by/>Operation and procedureActivating the function, introducing a certain perception offset for the input characteristic diagram, so that the input characteristic diagram is more convenient to train, and the output of the convolution module is recorded as/>The formula is:
Then to Respectively carry out/>, on two branchesAnd/>And splice the results through/>And restoring the channel number of the spliced result to be consistent with the input characteristic diagram, wherein the channel number is expressed as follows:
In the above, for The calculation mode is as follows:
wherein by inputting the feature map Global averaging pooling to obtain vector/>Its length is characteristic diagram/>Channel number/>. Subsequently, two full connection layers/>, are usedAnd/>Will/>Length transformation of/>2 Reconverted back/>Acting as a weight on the input feature map/>And the self-adaptive suppression of redundant characteristics is realized, and the important characteristics are enhanced.
And then the obtainedFeature map input/>The addition proceeds to the next stage of the decoder as a weighted feature.
Meanwhile, considering that part of the features are difficult to extract, additional weighting processing is needed, and a sparse channel attention module is added in the embodiment. Since the common class of images have highly correlated channel sparsity, the sparsity pattern of the channels contains discriminant information, and the sparsity of the feature map is used to show the importance of infrequently occurring features, embodiments obtain sparsity by computing the properties of the non-zero elements of each channel, for the input feature mapIs the kth feature map/>The sparsity is calculated as follows:
wherein, Indicating the sparsity of the kth channel. To increase the channel weight of the infrequent occurrence of features, the channel weight is redefined as:
wherein, Channel number of feature map,/>Is a minimum to ensure non-zero, and the examples set it to 0.0001 in the experiments.
The embodiment uses the calculation mode based on the channel feature sparsity to replace a global average pooling part in the SE module to obtain a weight vector capable of reflecting the feature channel sparsity and acts on the input feature map.
Finally, the embodiment splices the weighted feature decoding output and the space detail information obtained by the space detail feature extraction module, and then carries out semantic segmentation prediction through the segmentation head to obtain a marine remote sensing image segmentation result.
In addition, in the process of training to obtain a fully trained marine remote sensing image semantic segmentation network, the model total loss function value L comprises a cross entropy loss functionDiceLoss loss function/>Two parts, the formula is as follows:
wherein, And/>Representing the weight parameters.
Cross entropy lossThe calculation formula is as follows:
DiceLoss loss of The calculation formula is as follows:
wherein, For predictive semantic segmentation map,/>Ground Truth,/>, for semantic segmentationIs the category number.
In conclusion, multiscale semantic information in the ocean remote sensing image is extracted through crisscross attention, redundant features and enhanced detail features are suppressed through double-branch channel attention weighting, fine multiscale feature fusion is achieved, shallow space detail information and deep global context information are fused through cascade multiscale fusion, and accurate segmentation of ocean remote sensing image semantics is achieved.
Based on the semantic segmentation method of the marine remote sensing image provided by the invention, the invention also provides a semantic segmentation device 600 of the marine remote sensing image, as shown in fig. 6, comprising:
an image acquisition unit 601, configured to acquire a marine remote sensing image to be identified;
The semantic segmentation unit 602 is configured to input the marine remote sensing image to be identified into a marine remote sensing image semantic segmentation network with complete training, perform global context information extraction on the marine remote sensing image to be identified by using the marine remote sensing image semantic segmentation network to obtain a multi-scale feature map, perform space detail feature extraction on the multi-scale feature map to obtain space detail information, perform cascade multi-scale fusion, feature decoding and dual-branch channel attention weighting on the multi-scale feature map to obtain feature decoding output, and perform semantic segmentation prediction after the feature decoding output and the space detail information are spliced to obtain a marine remote sensing image segmentation result.
The semantic segmentation device 600 for ocean remote sensing images provided in the foregoing embodiment may implement the technical solution in the foregoing embodiment of the semantic segmentation method for ocean remote sensing images, and the specific implementation principle of each module or unit may refer to the corresponding content in the foregoing embodiment of the semantic segmentation method for ocean remote sensing images, which is not described herein again.
The present invention also provides an electronic device 700, as shown in fig. 7, fig. 7 is a schematic structural diagram of an embodiment of the electronic device provided by the present invention, where the electronic device 700 includes a processor 701, a memory 702, and a computer program stored in the memory 702 and capable of running on the processor 701, and when the processor 701 executes the program, the above-mentioned semantic segmentation method for ocean remote sensing images is implemented.
As a preferred embodiment, the electronic device further comprises a display 703 for displaying the process of executing the above-mentioned marine remote sensing image semantic segmentation method by the processor 701.
The processor 701 may be an integrated circuit chip, and has signal processing capability. The processor 701 may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), etc.; but also digital signal processors (DIGITAL SIGNAL processors, DSP), application SPECIFIC INTEGRATED Circuits (ASIC). The disclosed methods, steps, and logic blocks in the embodiments of the present invention may be implemented or performed. A general purpose processor may also be a microprocessor or the processor may be any conventional processor or the like.
The Memory 702 may be, but is not limited to, a random access Memory (Random Access Memory, RAM), a Read Only Memory (ROM), a Secure Digital (SD card), a flash Memory card (FLASH CARD), etc. The memory 702 is configured to store a program, and the processor 701 executes the program after receiving an execution instruction, and the method for defining a flow disclosed in any of the foregoing embodiments of the present invention may be applied to the processor 701 or implemented by the processor 701.
The display 703 may be an LED display screen, a liquid crystal display, a touch display, or the like. The display 703 is used to display various information on the electronic device 700.
It is to be appreciated that the configuration shown in fig. 7 is merely a schematic diagram of one configuration of the electronic device 700, and that the electronic device 700 may include more or fewer components than shown in fig. 7. The components shown in fig. 7 may be implemented in hardware, software, or a combination thereof.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention.
Claims (6)
1. A semantic segmentation method for marine remote sensing images, the method comprising:
acquiring a marine remote sensing image to be identified;
Inputting the ocean remote sensing image to be identified into a ocean remote sensing image semantic segmentation network with complete training, carrying out patch embedding on the ocean remote sensing image to be identified by the ocean remote sensing image semantic segmentation network to obtain a preliminary feature image, carrying out multi-stage cross feature extraction on the preliminary feature image step by step to obtain a multi-scale feature image, wherein the multi-scale feature image comprises a first feature image, a second feature image, a third feature image and a fourth feature image, each stage in the multi-stage cross feature extraction comprises a plurality of feature extraction modules, each feature extraction module comprises a normalization layer, cross multi-head attention, jump connection and multi-layer perception operation, carrying out space detail feature extraction on the multi-scale feature image to obtain space detail information, carrying out cascading multi-scale fusion, feature decoding and double-branch channel attention weighting on the multi-scale feature image to obtain feature decoding output, and carrying out semantic segmentation prediction after the feature decoding output and the space detail information are spliced to obtain a ocean remote sensing image segmentation result;
The step of performing cascading multi-scale fusion, feature decoding and dual-branch channel attention weighting on the multi-scale feature map to obtain feature decoding output comprises the following steps: performing self-adaptive downsampling and feature fusion on the multi-scale feature map to obtain a fusion feature map corresponding to each decoding stage, performing feature decoding on the fourth feature map after being spliced with the corresponding fusion feature map step by step to obtain feature decoding maps of all stages, performing preliminary convolution processing on the feature decoding maps, performing weight extraction on the feature map after the preliminary convolution processing based on SE channel attention to obtain feature weights, performing preliminary feature weighting on the feature map after the preliminary convolution processing based on the feature weights, performing sparsity calculation on the feature map after the preliminary convolution processing based on sparse channel attention to obtain channel weights, and performing channel feature weighting on the feature map after the preliminary convolution processing based on the channel weights to obtain feature decoding output.
2. The semantic segmentation method of marine remote sensing images according to claim 1, wherein the extracting spatial detail features from the multi-scale feature map to obtain spatial detail information comprises:
taking the first feature map as feature input for extracting first space detail features, taking the second feature map as guide input for extracting the first space detail features, and extracting the space detail features based on residual convolution to obtain first detail features;
Taking the first detail feature as a feature input for extracting a second space detail feature, taking the third feature map as a guide input for extracting the second space detail feature, and extracting the space detail feature based on residual convolution to obtain the second detail feature;
and taking the second detail feature as a feature input for extracting a third space detail feature, taking the fourth feature map as a guide input for extracting the third space detail feature, and extracting the space detail feature based on residual convolution to obtain the space detail feature.
3. The semantic segmentation method of marine remote sensing images according to claim 2, wherein the spatial detail feature extraction based on residual convolution comprises:
performing up-sampling and convolution mapping operation on the guiding input to obtain guiding characteristics;
And splicing the guide feature into the feature input, and carrying out convolution, normalization, batch regularization and activation functions twice in succession to obtain a detail feature output.
4. The semantic segmentation method of marine remote sensing images according to claim 1, wherein the performing adaptive downsampling and feature fusion on the multi-scale feature map to obtain a fused feature map corresponding to each decoding stage comprises:
determining the corresponding downsampling times of each multi-scale feature map according to the scale ordering of the multi-scale feature map;
Performing downsampling operation for the multi-scale feature map step by step for corresponding times according to the corresponding downsampling times to obtain multi-stage downsampling features, wherein the multi-stage downsampling features correspond to each decoding stage;
And fusing the multi-level downsampling characteristics corresponding to the same decoding stage, which are obtained by downsampling different multi-scale characteristic diagrams, to obtain the fused characteristic diagram corresponding to each decoding stage.
5. A semantic segmentation device for marine remote sensing images, comprising:
the image acquisition unit is used for acquiring the ocean remote sensing image to be identified;
The semantic segmentation unit is used for inputting the ocean remote sensing image to be identified into a ocean remote sensing image semantic segmentation network with complete training, the ocean remote sensing image semantic segmentation network patches the ocean remote sensing image to be identified to obtain a preliminary feature image, multi-stage crisscross feature extraction is carried out on the preliminary feature image step by step to obtain a multi-scale feature image, the multi-scale feature image comprises a first feature image, a second feature image, a third feature image and a fourth feature image, each stage in the multi-stage crisscross feature extraction comprises a plurality of feature extraction modules, each feature extraction module comprises a normalization layer, crisscross multi-head attention, jump connection and multi-layer perception operation, space detail feature extraction is carried out on the multi-scale feature image to obtain space detail information, cascading multi-scale fusion, feature decoding and ocean attention weighting of double branch channels are carried out on the multi-scale feature image to obtain feature decoding output, and semantic segmentation prediction is carried out after the feature decoding output and the space detail information are spliced to obtain a remote sensing image segmentation result;
The step of performing cascading multi-scale fusion, feature decoding and dual-branch channel attention weighting on the multi-scale feature map to obtain feature decoding output comprises the following steps: performing self-adaptive downsampling and feature fusion on the multi-scale feature map to obtain a fusion feature map corresponding to each decoding stage, performing feature decoding on the fourth feature map after being spliced with the corresponding fusion feature map step by step to obtain feature decoding maps of all stages, performing preliminary convolution processing on the feature decoding maps, performing weight extraction on the feature map after the preliminary convolution processing based on SE channel attention to obtain feature weights, performing preliminary feature weighting on the feature map after the preliminary convolution processing based on the feature weights, performing sparsity calculation on the feature map after the preliminary convolution processing based on sparse channel attention to obtain channel weights, and performing channel feature weighting on the feature map after the preliminary convolution processing based on the channel weights to obtain feature decoding output.
6. An electronic device comprising a memory and a processor, wherein,
The memory is used for storing a computer program;
The processor, coupled to the memory, is configured to execute a computer program to implement the steps in the marine remote sensing image semantic segmentation method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410015646.0A CN117522884B (en) | 2024-01-05 | 2024-01-05 | Ocean remote sensing image semantic segmentation method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410015646.0A CN117522884B (en) | 2024-01-05 | 2024-01-05 | Ocean remote sensing image semantic segmentation method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117522884A CN117522884A (en) | 2024-02-06 |
| CN117522884B true CN117522884B (en) | 2024-05-17 |
Family
ID=89751705
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410015646.0A Active CN117522884B (en) | 2024-01-05 | 2024-01-05 | Ocean remote sensing image semantic segmentation method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117522884B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117934518B (en) * | 2024-03-21 | 2024-06-04 | 国家海洋局南海规划与环境研究院 | Remote sensing image segmentation method and system for ocean homeland space detailed planning |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112287940A (en) * | 2020-10-30 | 2021-01-29 | 西安工程大学 | Semantic segmentation method of attention mechanism based on deep learning |
| CN113723411A (en) * | 2021-06-18 | 2021-11-30 | 湖北工业大学 | Feature extraction method and segmentation system for semantic segmentation of remote sensing image |
| CN114445430A (en) * | 2022-04-08 | 2022-05-06 | 暨南大学 | Real-time image semantic segmentation method and system for lightweight multi-scale feature fusion |
| CN114863179A (en) * | 2022-05-18 | 2022-08-05 | 合肥工业大学 | Endoscope image classification method based on multi-scale feature embedding and cross attention |
| CN115601549A (en) * | 2022-12-07 | 2023-01-13 | 山东锋士信息技术有限公司（Cn） | River and lake remote sensing image segmentation method based on deformable convolution and self-attention model |
| CN116665065A (en) * | 2023-07-28 | 2023-08-29 | 山东建筑大学 | Cross attention-based high-resolution remote sensing image change detection method |
| CN117237320A (en) * | 2023-06-19 | 2023-12-15 | 重庆邮电大学 | Multi-context brain tumor segmentation system based on scale fusion guidance |
| CN117274608A (en) * | 2023-11-23 | 2023-12-22 | 太原科技大学 | Remote sensing image semantic segmentation method based on space detail perception and attention guidance |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114627389B (en) * | 2022-03-23 | 2023-01-31 | 中国科学院空天信息创新研究院 | Raft culture area extraction method based on multi-temporal optical remote sensing image |
-
2024
- 2024-01-05 CN CN202410015646.0A patent/CN117522884B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112287940A (en) * | 2020-10-30 | 2021-01-29 | 西安工程大学 | Semantic segmentation method of attention mechanism based on deep learning |
| CN113723411A (en) * | 2021-06-18 | 2021-11-30 | 湖北工业大学 | Feature extraction method and segmentation system for semantic segmentation of remote sensing image |
| CN114445430A (en) * | 2022-04-08 | 2022-05-06 | 暨南大学 | Real-time image semantic segmentation method and system for lightweight multi-scale feature fusion |
| CN114863179A (en) * | 2022-05-18 | 2022-08-05 | 合肥工业大学 | Endoscope image classification method based on multi-scale feature embedding and cross attention |
| CN115601549A (en) * | 2022-12-07 | 2023-01-13 | 山东锋士信息技术有限公司（Cn） | River and lake remote sensing image segmentation method based on deformable convolution and self-attention model |
| CN117237320A (en) * | 2023-06-19 | 2023-12-15 | 重庆邮电大学 | Multi-context brain tumor segmentation system based on scale fusion guidance |
| CN116665065A (en) * | 2023-07-28 | 2023-08-29 | 山东建筑大学 | Cross attention-based high-resolution remote sensing image change detection method |
| CN117274608A (en) * | 2023-11-23 | 2023-12-22 | 太原科技大学 | Remote sensing image semantic segmentation method based on space detail perception and attention guidance |
Non-Patent Citations (1)
| Title |
|---|
| 基于多级通道注意力的遥感图像分割方法;余帅 等;激光与光电子学进展;20201231(第04期);142-151 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117522884A (en) | 2024-02-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | FE-YOLOv5: Feature enhancement network based on YOLOv5 for small object detection | |
| CN117522884B (en) | Ocean remote sensing image semantic segmentation method and device and electronic equipment | |
| CN111480169B (en) | Method, system and device for pattern recognition | |
| CN111598112B (en) | Multitask target detection method and device, electronic equipment and storage medium | |
| EP4053718A1 (en) | Watermark information embedding method and apparatus | |
| CN112084923A (en) | Semantic segmentation method for remote sensing image, storage medium and computing device | |
| CN115601661A (en) | Building change detection method for urban dynamic monitoring | |
| CN111160312A (en) | Target identification method and device and electronic equipment | |
| CN115293992A (en) | Polarization image defogging method and device based on unsupervised weight depth model | |
| CN113569607A (en) | Motion recognition method, motion recognition device, motion recognition equipment and storage medium | |
| CN115330703A (en) | Remote sensing image cloud and cloud shadow detection method based on context information fusion | |
| CN112132867B (en) | Remote sensing image change detection method and device | |
| CN114005120A (en) | License plate character cutting method, license plate recognition method, device, equipment and storage medium | |
| CN117649581A (en) | Image forgery detection method and device with multi-task edge enhancement | |
| CN117689617A (en) | Insulator detection method based on defogging constraint network and series connection multi-scale attention | |
| CN116258756B (en) | Self-supervision monocular depth estimation method and system | |
| US12112524B2 (en) | Image augmentation method, electronic device and readable storage medium | |
| CN116612130A (en) | Smoke segmentation method, system, equipment and medium based on deep neural network | |
| CN115329813A (en) | Improved U-Net based vertical ionogram O wave and X wave signal separation method | |
| CN114359633A (en) | Hyperspectral image clustering method and device, electronic equipment and storage medium | |
| CN115205624A (en) | Cross-dimension attention-convergence cloud and snow identification method and equipment and storage medium | |
| CN115240163A (en) | Traffic sign detection method and system based on one-stage detection network | |
| CN107748875A (en) | A kind of earthquake building recognition method based on multidate radar image texture feature | |
| CN115147434A (en) | Image processing method, device, terminal equipment and computer readable storage medium | |
| CN112052863A (en) | Image detection method and device, computer storage medium and electronic equipment |
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 |