CN113920407A - Ship target identification method and system based on multiband remote sensing image fusion - Google Patents
Ship target identification method and system based on multiband remote sensing image fusion Download PDFInfo
- Publication number
- CN113920407A CN113920407A CN202111001651.9A CN202111001651A CN113920407A CN 113920407 A CN113920407 A CN 113920407A CN 202111001651 A CN202111001651 A CN 202111001651A CN 113920407 A CN113920407 A CN 113920407A
- Authority
- CN
- China
- Prior art keywords
- image
- network
- ship
- fusion
- multiband
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The invention discloses a ship target identification method and a system based on multiband remote sensing image fusion, wherein the method comprises the following steps: acquiring a multiband remote sensing image subjected to multiband registration, extracting the characteristics of the remote sensing image of a plurality of wavebands through a convolutional neural network, continuously updating pixels of a fused image through continuous reverse propagation based on a generative model, and outputting the fused image to a ship target detection step after the optimal effect is achieved; receiving the fused image output by the multiband image fusion step, finishing the detection of the ship target in the fused image based on an end-to-end target detection network, and outputting a slice image of the ship target and a rough classification result of the ship category to a ship target identification step; and acquiring the slice image of the ship target output in the ship target detection step, and outputting a final identification result of the ship type after a refined discrimination network. The invention strengthens the characteristics of the ship target and makes up the defect that single-waveband imaging is easy to be interfered.
Description
Technical Field
The invention relates to a ship target identification method and system based on multiband remote sensing image fusion, and belongs to the technical field of multiband image fusion technology, target detection and target identification.
Background
With the continuous progress of the remote sensing technology, the quality of the optical remote sensing image is also continuously improved, the remote sensing image is reasonably and fully utilized, and the method not only contributes to civil aspects such as land resource management, urban construction planning and ecological environment protection, but also helps military aspects. The remote sensing image has abundant surface feature texture information, so that the target identification of the remote sensing image is an important problem in the field of aviation and satellite image analysis. In the ship target identification technology, due to uncertain conditions such as atmosphere, illumination, cloud layer and islands on the surface of the ocean, the traditional ship identification method has low identification precision and poor stability, so that the research on a high-precision and high-stability ship identification technology has very important value.
The existing optical remote sensing image ship target identification technology, especially the research work developed based on deep learning, adopts the following methods: the method comprises the steps of directly utilizing a convolutional neural network to extract depth features of an input optical remote sensing image, constructing a classifier, and utilizing the features to carry out classification and regression to obtain circumscribed rectangular coordinates of a region where a ship target is located and the category of the ship target.
Disclosure of Invention
Most of existing optical remote sensing image ship target identification methods are based on single-waveband images, the available information is relatively few, the ship identification accuracy is low, the ship identification method is prone to interference of factors such as illumination and cloud and fog, and the ship identification robustness is low. The invention aims to overcome the technical defects in the prior art, solve the technical problems and provide a ship target identification method and system based on multiband remote sensing image fusion.
The invention specifically adopts the following technical scheme: a ship target identification method based on multiband remote sensing image fusion comprises the following steps:
the multi-band image fusion step specifically comprises the following steps: acquiring a multiband remote sensing image subjected to multiband registration, extracting the characteristics of the remote sensing image of a plurality of wavebands through a convolutional neural network, continuously updating pixels of a fused image through continuous reverse propagation based on a generative model, and outputting the fused image to a ship target detection step after the optimal effect is achieved;
the ship target detection method specifically comprises the following steps: receiving the fused image output by the multiband image fusion step, finishing the detection of the ship target in the fused image based on an end-to-end target detection network, and outputting a slice image of the ship target and a rough classification result of the ship category to a ship target identification step;
the ship target identification step specifically comprises the following steps: and acquiring the slice image of the ship target output in the ship target detection step, and outputting a final identification result of the ship type after a refined discrimination network.
As a preferred embodiment, the multi-band image fusion step includes: the fusion of multiband remote sensing images is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network and a fusion network, and the method specifically comprises the following steps:
firstly, registering input multiband remote sensing image data under the same time phase to obtain each waveband image after pixel alignment;
then, completing depth feature extraction of each band image through the feature extraction network, wherein the feature extraction network is a depth convolution neural network, and the input image can output a feature map after being convolved;
and secondly, fusing the characteristic graphs of all the wave bands through a fusion network to finally generate a fused image.
In a preferred embodiment, the fusion network is a generation network, the feature map of the multiband image and a random noise image are initially input, the pixel values of the noise image are continuously updated by using back propagation, the pixel updating of the noise image is controlled by a fusion loss function, the loss is reduced to the minimum after a plurality of iterations, the network completes the pixel updating, and a new image, namely an output fused image, is generated.
As a preferred embodiment, the ship detecting step includes: the detection of the ship target in the multiband remote sensing fusion image is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network, a region recommendation network, a feature alignment network and a classification regression network, and the method specifically comprises the following steps:
firstly, inputting fused multiband remote sensing images into a feature extraction network, wherein the feature extraction network performs convolution on the images to extract high-level features of the fused images;
then, inputting the extracted high-level features into a regional recommendation network, predicting by the regional recommendation network by using a high-level feature map, and outputting a target recommendation region frame;
then, inputting the recommended region box of the target into a feature alignment network for feature alignment;
and finally, inputting the features after feature alignment into a classification regression network, correcting a ship target frame, predicting a target class, and finally outputting a corrected ship target position and a rough classification result, namely confidence degrees of the target frame and each class corresponding to the target frame.
As a preferred embodiment, the ship identifying step specifically includes:
firstly, after the output of the ship detection step is sliced, the slice image of each ship is used as the input image of the ship identification step;
then, extracting the depth features of the slice images by using a feature extraction network, sending the last layer of feature images extracted by the feature extraction network into a classifier 1 for classification, simultaneously extracting a plurality of feature images under different scales and sending the feature images into a multi-scale branch, and classifying the multi-scale feature images in the multi-scale branch by using a classifier 2 to obtain a classification result under multiple scales;
and finally, fusing the classification results output by the classifier 1 and the classifier 2 and the rough classification result of the ship detection step by using a decision-level fusion algorithm, further optimizing the confidence of each type, and finally outputting the type with the highest confidence as the output result of the ship identification step.
The invention also provides a ship target identification system based on multiband remote sensing image fusion, which comprises:
the multiband image fusion module specifically executes: acquiring a multiband remote sensing image subjected to multiband registration, extracting the characteristics of the remote sensing image of a plurality of wavebands through a convolutional neural network, continuously updating the pixels of a fused image through continuous reverse propagation based on a generative model, and outputting the fused image to a ship target detection module after the optimal effect is achieved;
the ship target detection module specifically executes: receiving the fused image output by the multiband image fusion module, finishing the detection of the ship target in the fused image based on an end-to-end target detection network, and outputting a slice image of the ship target and a rough classification result of the ship category to a ship target identification module;
the ship target identification module specifically executes: and acquiring the slice image of the ship target output by the ship target detection module, and outputting a final identification result of the ship type after a refined discrimination network.
As a preferred embodiment, the multiband image fusion module specifically performs: the fusion of multiband remote sensing images is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network and a fusion network, and the method specifically comprises the following steps:
firstly, registering input multiband remote sensing image data under the same time phase to obtain each waveband image after pixel alignment;
then, completing depth feature extraction of each band image through the feature extraction network, wherein the feature extraction network is a depth convolution neural network, and the input image can output a feature map after being convolved;
and secondly, fusing the characteristic graphs of all the wave bands through a fusion network to finally generate a fused image.
In a preferred embodiment, the fusion network is a generation network, the feature map of the multiband image and a random noise image are initially input, the pixel values of the noise image are continuously updated by using back propagation, the pixel updating of the noise image is controlled by a fusion loss function, the loss is reduced to the minimum after a plurality of iterations, the network completes the pixel updating, and a new image, namely an output fused image, is generated.
As a preferred embodiment, the ship detection module specifically performs: the detection of the ship target in the multiband remote sensing fusion image is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network, a region recommendation network, a feature alignment network and a classification regression network, and the method specifically comprises the following steps:
firstly, inputting fused multiband remote sensing images into a feature extraction network, wherein the feature extraction network performs convolution on the images to extract high-level features of the fused images;
then, inputting the extracted high-level features into a regional recommendation network, predicting by the regional recommendation network by using a high-level feature map, and outputting a target recommendation region frame;
then, inputting the recommended region box of the target into a feature alignment network for feature alignment;
and finally, inputting the features after feature alignment into a classification regression network, correcting a ship target frame, predicting a target class, and finally outputting a corrected ship target position and a rough classification result, namely confidence degrees of the target frame and each class corresponding to the target frame.
As a preferred embodiment, the ship identification module specifically performs:
firstly, after the output of a ship detection module is sliced, a slice image of each ship is used as an input image of a ship identification module;
then, extracting the depth features of the slice images by using a feature extraction network, sending the last layer of feature images extracted by the feature extraction network into a classifier 1 for classification, simultaneously extracting a plurality of feature images under different scales and sending the feature images into a multi-scale branch, and classifying the multi-scale feature images in the multi-scale branch by using a classifier 2 to obtain a classification result under multiple scales;
and finally, fusing the classification results output by the classifier 1 and the classifier 2 and the rough classification result of the ship detection module by using a decision-level fusion algorithm, further optimizing the confidence of each type, and finally outputting the type with the highest confidence as the output result of the ship identification module.
The invention achieves the following beneficial effects: firstly, the whole system is composed of a plurality of modules, a fusion module fuses multiband images, a ship detection module detects a ship target in the fused images, and finally a ship identification module completes identification. Secondly, the method and the system of the invention fuse the characteristics of the remote sensing images with a plurality of wave bands, achieve the purpose of getting strong points and compensating weak points, and enrich the available information of the ship target. Thirdly, the method and the system adopt a technical route of firstly detecting the ship target and then finely distinguishing the target, so that the accuracy of identification is improved while high recall rate is ensured. Fourthly, the method and the system provided by the invention adopt a decision-level fusion method to fuse the coarse classification result, the fine classification result and the multi-scale classification result, thereby further improving the identification precision.
Drawings
FIG. 1 is a schematic diagram of a topological architecture of a ship target identification system based on multiband remote sensing image fusion according to the invention;
FIG. 2 is a schematic of the topology of the multi-band image fusion module of the present invention;
FIG. 3 is a schematic of the topology of the ship detection module of the present invention;
FIG. 4 is a schematic of the topology of the ship identification module of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
Example 1: the invention provides a ship target identification method based on multiband remote sensing image fusion, which comprises the following steps:
the multi-band image fusion step specifically comprises the following steps: acquiring a multiband remote sensing image subjected to multiband registration, extracting the characteristics of the remote sensing image of a plurality of wavebands through a convolutional neural network, continuously updating pixels of a fused image through continuous reverse propagation based on a generative model, and outputting the fused image to a ship target detection step after the optimal effect is achieved;
the ship target detection method specifically comprises the following steps: receiving the fused image output by the multiband image fusion step, finishing the detection of the ship target in the fused image based on an end-to-end target detection network, and outputting a slice image of the ship target and a rough classification result of the ship category to a ship target identification step;
the ship target identification step specifically comprises the following steps: and acquiring the slice image of the ship target output in the ship target detection step, and outputting a final identification result of the ship type after a refined discrimination network.
As a preferred embodiment, the multi-band image fusion step includes: the fusion of multiband remote sensing images is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network and a fusion network, and the method specifically comprises the following steps:
firstly, registering input multiband remote sensing image data under the same time phase to obtain each waveband image after pixel alignment;
then, completing depth feature extraction of each band image through the feature extraction network, wherein the feature extraction network is a depth convolution neural network, and the input image can output a feature map after being convolved;
and secondly, fusing the characteristic graphs of all the wave bands through a fusion network to finally generate a fused image.
In a preferred embodiment, the fusion network is a generation network, the feature map of the multiband image and a random noise image are initially input, the pixel values of the noise image are continuously updated by using back propagation, the pixel updating of the noise image is controlled by a fusion loss function, the loss is reduced to the minimum after a plurality of iterations, the network completes the pixel updating, and a new image, namely an output fused image, is generated.
As a preferred embodiment, the ship detecting step includes: the detection of the ship target in the multiband remote sensing fusion image is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network, a region recommendation network, a feature alignment network and a classification regression network, and the method specifically comprises the following steps:
firstly, inputting fused multiband remote sensing images into a feature extraction network, wherein the feature extraction network performs convolution on the images to extract high-level features of the fused images;
then, inputting the extracted high-level features into a regional recommendation network, predicting by the regional recommendation network by using a high-level feature map, and outputting a target recommendation region frame;
then, inputting the recommended region box of the target into a feature alignment network for feature alignment;
and finally, inputting the features after feature alignment into a classification regression network, correcting a ship target frame, predicting a target class, and finally outputting a corrected ship target position and a rough classification result, namely confidence degrees of the target frame and each class corresponding to the target frame.
As a preferred embodiment, the ship identifying step specifically includes:
firstly, after the output of the ship detection step is sliced, the slice image of each ship is used as the input image of the ship identification step;
then, extracting the depth features of the slice images by using a feature extraction network, sending the last layer of feature images extracted by the feature extraction network into a classifier 1 for classification, simultaneously extracting a plurality of feature images under different scales and sending the feature images into a multi-scale branch, and classifying the multi-scale feature images in the multi-scale branch by using a classifier 2 to obtain a classification result under multiple scales;
and finally, fusing the classification results output by the classifier 1 and the classifier 2 and the rough classification result of the ship detection step by using a decision-level fusion algorithm, further optimizing the confidence of each type, and finally outputting the type with the highest confidence as the output result of the ship identification step.
Example 2: as shown in fig. 1, the present invention further provides a ship target identification system based on multiband remote sensing image fusion, including:
the multiband image fusion module specifically executes: acquiring a multiband remote sensing image subjected to multiband registration, extracting the characteristics of the remote sensing image of a plurality of wavebands through a convolutional neural network, continuously updating the pixels of a fused image through continuous reverse propagation based on a generative model, and outputting the fused image to a ship target detection module after the optimal effect is achieved;
the ship target detection module specifically executes: receiving the fused image output by the multiband image fusion module, finishing the detection of the ship target in the fused image based on an end-to-end target detection network, and outputting a slice image of the ship target and a rough classification result of the ship category to a ship target identification module;
the ship target identification module specifically executes: and acquiring the slice image of the ship target output by the ship target detection module, and outputting a final identification result of the ship type after a refined discrimination network.
As a preferred embodiment, as shown in fig. 2, the multi-band image fusion module specifically performs: the fusion of multiband remote sensing images is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network and a fusion network, and the method specifically comprises the following steps:
firstly, registering input multiband remote sensing image data under the same time phase to obtain each waveband image after pixel alignment;
then, completing depth feature extraction of each band image through the feature extraction network, wherein the feature extraction network is a depth convolution neural network, and the input image can output a feature map after being convolved;
and secondly, fusing the characteristic graphs of all the wave bands through a fusion network to finally generate a fused image.
In a preferred embodiment, the fusion network is a generation network, the feature map of the multiband image and a random noise image are initially input, the pixel values of the noise image are continuously updated by using back propagation, the pixel updating of the noise image is controlled by a fusion loss function, the loss is reduced to the minimum after a plurality of iterations, the network completes the pixel updating, and a new image, namely an output fused image, is generated.
As shown in fig. 3, as a preferred embodiment, the ship detection module specifically performs: the detection of the ship target in the multiband remote sensing fusion image is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network, a region recommendation network, a feature alignment network and a classification regression network, and the method specifically comprises the following steps:
firstly, inputting fused multiband remote sensing images into a feature extraction network, wherein the feature extraction network performs convolution on the images to extract high-level features of the fused images;
then, inputting the extracted high-level features into a regional recommendation network, predicting by the regional recommendation network by using a high-level feature map, and outputting a target recommendation region frame;
then, inputting the recommended region box of the target into a feature alignment network for feature alignment;
and finally, inputting the features after feature alignment into a classification regression network, correcting a ship target frame, predicting a target class, and finally outputting a corrected ship target position and a rough classification result, namely confidence degrees of the target frame and each class corresponding to the target frame.
As shown in fig. 4, as a preferred embodiment, the ship identification module specifically executes:
firstly, after the output of a ship detection module is sliced, a slice image of each ship is used as an input image of a ship identification module;
then, extracting the depth features of the slice images by using a feature extraction network, sending the last layer of feature images extracted by the feature extraction network into a classifier 1 for classification, simultaneously extracting a plurality of feature images under different scales and sending the feature images into a multi-scale branch, and classifying the multi-scale feature images in the multi-scale branch by using a classifier 2 to obtain a classification result under multiple scales;
and finally, fusing the classification results output by the classifier 1 and the classifier 2 and the rough classification result of the ship detection module by using a decision-level fusion algorithm, further optimizing the confidence of each type, and finally outputting the type with the highest confidence as the output result of the ship identification module.
The key points of the invention are as follows: firstly, the whole method is composed of a plurality of modules, a fusion module fuses multiband images, the fused images are detected by a ship target of a ship detection module, and finally, the ship identification module is used for completing identification. Secondly, the fusion module extracts the depth characteristics of the multiband remote sensing image by using the convolutional neural network, then performs fusion by using a generating network, and controls the fusion process by using a loss function, so that the full automation of the fusion is realized, and the fusion effect is improved. Thirdly, in the ship identification process, the ship target position and the rough classification result are obtained through detection, then refined judgment is carried out, the multi-scale features and the rough classification result are combined in the identification process, and decision-level fusion is carried out, so that the ship identification precision is further improved.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.
Claims (10)
1. A ship target identification method based on multiband remote sensing image fusion is characterized by comprising the following steps:
the multi-band image fusion step specifically comprises the following steps: acquiring a multiband remote sensing image subjected to multiband registration, extracting the characteristics of the remote sensing image of a plurality of wavebands through a convolutional neural network, continuously updating pixels of a fused image through continuous reverse propagation based on a generative model, and outputting the fused image to a ship target detection step after the optimal effect is achieved;
the ship target detection method specifically comprises the following steps: receiving the fused image output by the multiband image fusion step, finishing the detection of the ship target in the fused image based on an end-to-end target detection network, and outputting a slice image of the ship target and a rough classification result of the ship category to a ship target identification step;
the ship target identification step specifically comprises the following steps: and acquiring the slice image of the ship target output in the ship target detection step, and outputting a final identification result of the ship type after a refined discrimination network.
2. The method for identifying the ship target based on the multiband remote sensing image fusion of claim 1, wherein the multiband image fusion step comprises: the fusion of multiband remote sensing images is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network and a fusion network, and the method specifically comprises the following steps:
firstly, registering input multiband remote sensing image data under the same time phase to obtain each waveband image after pixel alignment;
then, completing depth feature extraction of each band image through the feature extraction network, wherein the feature extraction network is a depth convolution neural network, and the input image can output a feature map after being convolved;
and secondly, fusing the characteristic graphs of all the wave bands through a fusion network to finally generate a fused image.
3. The method for identifying the ship target based on the fusion of the multiband remote sensing images as claimed in claim 2, wherein the fusion network is a generating network, the initial input is a feature map of the multiband images and a random noise image, the pixel values of the noise image are continuously updated by using back propagation, the pixel updating of the noise image is controlled by a fusion loss function, the loss is reduced to the minimum after a plurality of iterations, the network completes the pixel updating, and a new image, namely an output fused image, is generated.
4. The method for identifying the ship target based on the multiband remote sensing image fusion of claim 1, wherein the ship detection step comprises: the detection of the ship target in the multiband remote sensing fusion image is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network, a region recommendation network, a feature alignment network and a classification regression network, and the method specifically comprises the following steps:
firstly, inputting fused multiband remote sensing images into a feature extraction network, wherein the feature extraction network performs convolution on the images to extract high-level features of the fused images;
then, inputting the extracted high-level features into a regional recommendation network, predicting by the regional recommendation network by using a high-level feature map, and outputting a target recommendation region frame;
then, inputting the recommended region box of the target into a feature alignment network for feature alignment;
and finally, inputting the features after feature alignment into a classification regression network, correcting a ship target frame, predicting a target class, and finally outputting a corrected ship target position and a rough classification result, namely confidence degrees of the target frame and each class corresponding to the target frame.
5. The method for identifying the ship target based on the multiband remote sensing image fusion as claimed in claim 1, wherein the ship identification step specifically comprises:
firstly, after the output of the ship detection step is sliced, the slice image of each ship is used as the input image of the ship identification step;
then, extracting the depth features of the slice images by using a feature extraction network, sending the last layer of feature images extracted by the feature extraction network into a classifier 1 for classification, simultaneously extracting a plurality of feature images under different scales and sending the feature images into a multi-scale branch, and classifying the multi-scale feature images in the multi-scale branch by using a classifier 2 to obtain a classification result under multiple scales;
and finally, fusing the classification results output by the classifier 1 and the classifier 2 and the rough classification result of the ship detection step by using a decision-level fusion algorithm, further optimizing the confidence of each type, and finally outputting the type with the highest confidence as the output result of the ship identification step.
6. A ship target identification system based on multiband remote sensing image fusion is characterized by comprising:
the multiband image fusion module specifically executes: acquiring a multiband remote sensing image subjected to multiband registration, extracting the characteristics of the remote sensing image of a plurality of wavebands through a convolutional neural network, continuously updating the pixels of a fused image through continuous reverse propagation based on a generative model, and outputting the fused image to a ship target detection module after the optimal effect is achieved;
the ship target detection module specifically executes: receiving the fused image output by the multiband image fusion module, finishing the detection of the ship target in the fused image based on an end-to-end target detection network, and outputting a slice image of the ship target and a rough classification result of the ship category to a ship target identification module;
the ship target identification module specifically executes: and acquiring the slice image of the ship target output by the ship target detection module, and outputting a final identification result of the ship type after a refined discrimination network.
7. The system for identifying a ship target based on multiband remote sensing image fusion of claim 6, wherein the multiband image fusion module specifically executes: the fusion of multiband remote sensing images is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network and a fusion network, and the method specifically comprises the following steps:
firstly, registering input multiband remote sensing image data under the same time phase to obtain each waveband image after pixel alignment;
then, completing depth feature extraction of each band image through the feature extraction network, wherein the feature extraction network is a depth convolution neural network, and the input image can output a feature map after being convolved;
and secondly, fusing the characteristic graphs of all the wave bands through a fusion network to finally generate a fused image.
8. The system of claim 7, wherein the fusion network is a generating network, the initial inputs are a feature map of the multiband image and a random noise image, the pixel values of the noise image are continuously updated by using back propagation, the pixel update of the noise image is controlled by a fusion loss function, the loss is reduced to the minimum after a plurality of iterations, the network completes the pixel update, and a new image, namely an output fused image, is generated.
9. The system for identifying a ship target based on multiband remote sensing image fusion of claim 6, wherein the ship detection module specifically executes: the detection of the ship target in the multiband remote sensing fusion image is completed through a convolutional neural network framework, wherein the convolutional neural network comprises a feature extraction network, a region recommendation network, a feature alignment network and a classification regression network, and the method specifically comprises the following steps:
firstly, inputting fused multiband remote sensing images into a feature extraction network, wherein the feature extraction network performs convolution on the images to extract high-level features of the fused images;
then, inputting the extracted high-level features into a regional recommendation network, predicting by the regional recommendation network by using a high-level feature map, and outputting a target recommendation region frame;
then, inputting the recommended region box of the target into a feature alignment network for feature alignment;
and finally, inputting the features after feature alignment into a classification regression network, correcting a ship target frame, predicting a target class, and finally outputting a corrected ship target position and a rough classification result, namely confidence degrees of the target frame and each class corresponding to the target frame.
10. The system for identifying a ship target based on multiband remote sensing image fusion of claim 6, wherein the ship identification module specifically executes:
firstly, after the output of a ship detection module is sliced, a slice image of each ship is used as an input image of a ship identification module;
then, extracting the depth features of the slice images by using a feature extraction network, sending the last layer of feature images extracted by the feature extraction network into a classifier 1 for classification, simultaneously extracting a plurality of feature images under different scales and sending the feature images into a multi-scale branch, and classifying the multi-scale feature images in the multi-scale branch by using a classifier 2 to obtain a classification result under multiple scales;
and finally, fusing the classification results output by the classifier 1 and the classifier 2 and the rough classification result of the ship detection module by using a decision-level fusion algorithm, further optimizing the confidence of each type, and finally outputting the type with the highest confidence as the output result of the ship identification module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111001651.9A CN113920407A (en) | 2021-08-30 | 2021-08-30 | Ship target identification method and system based on multiband remote sensing image fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111001651.9A CN113920407A (en) | 2021-08-30 | 2021-08-30 | Ship target identification method and system based on multiband remote sensing image fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113920407A true CN113920407A (en) | 2022-01-11 |
Family
ID=79233473
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111001651.9A Pending CN113920407A (en) | 2021-08-30 | 2021-08-30 | Ship target identification method and system based on multiband remote sensing image fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113920407A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115546651A (en) * | 2022-11-17 | 2022-12-30 | 福建中科中欣智能科技有限公司 | Multimode ship target detection and recognition system and device |
-
2021
- 2021-08-30 CN CN202111001651.9A patent/CN113920407A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115546651A (en) * | 2022-11-17 | 2022-12-30 | 福建中科中欣智能科技有限公司 | Multimode ship target detection and recognition system and device |
| CN115546651B (en) * | 2022-11-17 | 2023-02-28 | 福建中科中欣智能科技有限公司 | Multimode ship target detection and recognition system and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112287807B (en) | Remote sensing image road extraction method based on multi-branch pyramid neural network | |
| CN111640159A (en) | Remote sensing image change detection method based on twin convolutional neural network | |
| CN110163213B (en) | Remote sensing image segmentation method based on disparity map and multi-scale depth network model | |
| CN109389163B (en) | Unmanned aerial vehicle image classification system and method based on topographic map | |
| CN110853026A (en) | Remote sensing image change detection method integrating deep learning and region segmentation | |
| CN107480727A (en) | The unmanned plane image fast matching method that a kind of SIFT and ORB are combined | |
| CN112837315B (en) | Deep learning-based transmission line insulator defect detection method | |
| CN109829423B (en) | Infrared imaging detection method for frozen lake | |
| CN109840553A (en) | The extracting method and system, storage medium, electronic equipment for agrotype of ploughing | |
| CN112884791B (en) | Method for constructing large-scale remote sensing image semantic segmentation model training sample set | |
| CN103544505A (en) | Ship recognition system and ship recognition method for aerial image pickup of unmanned plane | |
| CN111539432B (en) | Method for extracting urban road by using multi-source data to assist remote sensing image | |
| CN103871072A (en) | Method for automatic extraction of orthoimage embedding line based on projection digital elevation models | |
| CN111461006B (en) | Optical remote sensing image tower position detection method based on deep migration learning | |
| CN112434119A (en) | High-precision map production device based on heterogeneous data fusion | |
| CN111027511A (en) | Remote sensing image ship detection method based on region of interest block extraction | |
| CN116503705B (en) | Fusion method of digital city multi-source data | |
| CN112084871B (en) | High-resolution remote sensing target boundary extraction method based on weak supervised learning | |
| CN103839286A (en) | True-orthophoto optimization sampling method of object semantic constraint | |
| CN110956207A (en) | Optical remote sensing image full-element change detection method | |
| CN114708313A (en) | Optical and SAR image registration method based on double-branch neural network | |
| CN113920407A (en) | Ship target identification method and system based on multiband remote sensing image fusion | |
| CN112989985B (en) | Urban built-up area extraction method integrating night light data and Landsat8OLI images | |
| CN116091706B (en) | Three-dimensional reconstruction method for multi-mode remote sensing image deep learning matching | |
| CN114078106A (en) | Defect detection method based on improved Faster R-CNN |
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 |