Underwater target perception enhancement method based on target domain observation information guidance
By establishing attenuation feature semantic labels and generators in underwater target perception, the problem of unfit enhancement caused by the quality differences between the target domain and the source domain is solved, and the appropriate enhancement of underwater images under different attenuation conditions is achieved, which improves the underwater Perception quality in fields such as object detection and autonomous perception robots.
Patent Information
- Application Number
- CN202411116446.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-08-14
Smart Images

Figure CN118918456A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of target perception enhancement, and in particular relates to an underwater target perception enhancement method based on guidance of target domain observation information. Background Art
[0002] Under the supervision of domain prior knowledge or data information, the current algorithm has good perceptual quality enhancement performance in the source domain (training set). In practical applications, the diversity of the target domain (working environment) is an important factor affecting the performance and accuracy of visual tasks. When there is a quality difference between the source domain and the target domain, the same distribution assumption is not met, making it difficult for the enhancement model trained based on the source domain to act on the target domain, resulting in ill-posed enhancement results.
[0003] In order to improve the adaptability to underwater target domains with different attenuation conditions, the present invention proposes a method for enhancing the perceived quality of underwater targets based on target domain observation information. First, taking into account the differences in attenuation characteristics of different types of water bodies, semantic labels for different attenuation characteristics are established. Secondly, in view of the lack of high-dimensional semantic labels for the attenuation information in the target domain, an attenuation label generator and responder are established based on clustering technology to guide the enhancement model. Then, at the model level, enhancement modes that respond to different semantic labels are designed, and at the data level, reference truth values corresponding to different semantic labels are established to enrich the source domain information to respond to different attenuation characteristics. Finally, under the guidance of the target domain attenuation information, the enhancement method can significantly improve the cross-domain enhancement effect and improve the practical application value of the target domain enhanced image. Summary of the invention
[0004] 1. Technical issues to be resolved
[0005] The technical problem to be solved by the present invention is how to provide an underwater target perception enhancement method guided by target domain observation information to solve the ill-posed enhancement problem caused by the quality difference between the target domain and the source domain.
[0006] (II) Technical solution
[0007] In order to solve the above technical problems, the present invention proposes an underwater target perception enhancement method based on target domain observation information guidance, the method comprising the following steps:
[0008] S1. Attenuation feature establishment step: according to the attenuation feature differences of different types of water bodies, semantic labels of fuzzy attenuation, strong attenuation and multiple attenuation features are established;
[0009] S2, reference truth value generation step of different attenuation features: extract different features from the reference truth value data set to obtain reference truth values corresponding to different attenuation features, enriching the source domain information at the data level;
[0010] S3, low-quality data generation step: performing context feature degradation and texture feature degradation processing on the reference truth data set to obtain a low-quality data set;
[0011] S4, training data generation step: performing different feature extraction processes on the texture feature attenuated data set to obtain texture feature enhancement results corresponding to different attenuation features, which are used to extract features specific to the target scene from the data;
[0012] S5, multi-head decoding model training steps: During the training process, the feature extraction model and the multi-head decoding model are closely connected at the algorithm level by means of data enhancement. During the training process, the L1-norm is used as the loss function for training;
[0013] S6, step of establishing pseudo labels of attenuation features. Cluster the training data, establish the prior distribution of attenuation features, count the main attenuation features and generate pseudo labels to describe the attenuation features of the target domain observation input;
[0014] S7, observation information extraction step: establishing an attenuation feature generator to output an attenuation feature label according to the attenuation feature of the target domain;
[0015] S8, attenuation feature response step: Based on the multi-focus image fusion technology, an attenuation label responder is established to generate images and fuse significant texture features to output perceptual quality enhancement results.
[0016] (III) Beneficial effects
[0017] The present invention proposes an underwater target perception enhancement method based on the guidance of target domain observation information. The underwater target perception enhancement method based on the guidance of target domain observation information proposed by the present invention can solve the ill-posed enhancement problem caused by the quality difference between the target domain and the source domain. The present invention guides the source domain model to generate and fuse significant texture features and context features by taking the attenuation features of the target domain image as semantic labels, thereby realizing the well-posed enhancement of underwater images under different attenuation conditions, and has important application value in the fields of underwater target detection, key object perception, autonomous perception robots, etc.
[0018] The method proposed in the present invention takes the attenuation characteristics of the target domain image as semantic labels, which can guide the source domain enhancement model to generate images and fuse significant texture features, thereby achieving adaptive enhancement of the perception quality of underwater observation targets under different attenuation conditions, improving the operational performance of the visual system, and having important application value in the fields of underwater target detection, key object perception, and autonomous perception robots. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 The present invention is a construction step of guiding the underwater target perception quality enhancement method based on target domain observation information;
[0020] Figure 2 It is a model structure diagram of different attenuation feature extraction of the present invention;
[0021] Figure 3 It is a model structure diagram of a multi-head decoding of the present invention;
[0022] Figure 4 This is a model structure of an attenuation label generator of the present invention. DETAILED DESCRIPTION
[0023] In order to make the purpose, content and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the drawings and examples.
[0024] The present invention relates to a method for enhancing underwater target perception quality based on target domain observation information guidance. Specifically, in view of the ill-posed enhancement problem caused by the quality difference between the target domain and the source domain, a method is designed that can produce well-posed enhancement results in the target domain under different attenuation conditions according to the actual attenuation state.
[0025] The present invention discloses a method for enhancing the perception quality of underwater targets based on target domain observation information guidance, the method comprising: (1) an attenuation feature establishment step. According to the attenuation feature differences of different types of water bodies, semantic labels of fuzzy attenuation, strong attenuation and multiple attenuation features are established. (2) a reference truth value generation step for different attenuation features. The original reference truth value data set is subjected to different feature extraction to obtain reference truth values corresponding to different attenuation features, thereby enriching the source domain information at the data level. (3) a low-quality data generation step. The original reference truth value data set is subjected to context feature degradation and texture feature degradation processing to obtain a low-quality data set. (4) a training data generation step. The texture feature attenuated data set is subjected to different feature extraction processing to obtain texture feature enhancement results corresponding to different attenuation features, which are used to extract target scene-specific features from the data. (5) a multi-head decoding model training step. During the training process, the feature extraction model and the multi-head decoding model are closely connected at the algorithm level in a data enhancement manner to reduce the cumulative error. During the training process, the L1 norm is used as the loss function for joint optimization. (6) an attenuation feature pseudo label establishment step. The training data is clustered to establish the prior distribution of attenuation features. The main attenuation features are counted and pseudo labels are generated to describe the attenuation features of the target domain observation input. (7) Observation information extraction step. An attenuation feature generator is established to output semantic labels based on the attenuation features of the target domain. (8) Attenuation feature response step. An attenuation label responder is established to guide the enhancement model to generate images based on the semantic labels and fuse significant texture features to output the perceptual quality enhancement results.
[0026] The purpose of the present invention is to provide a method for enhancing the perception quality of underwater targets based on the guidance of target domain observation information, so as to solve the ill-posed enhancement problem caused by the quality difference between the target domain and the source domain. Under the guidance of the target domain observation information, the enhancement model can significantly improve the cross-domain enhancement effect, thereby improving the perception quality of the observed objects in the target domain.
[0027] To achieve the above object, the present invention proposes a method for enhancing underwater target perception quality based on target domain observation information guidance, the method comprising:
[0028] S1. Attenuation feature establishment step: according to the attenuation feature differences of different types of water bodies, semantic labels of fuzzy attenuation, strong attenuation and multiple attenuation features are established;
[0029] S2, reference truth value generation step of different attenuation features: extract different features from the reference truth value data set to obtain reference truth values corresponding to different attenuation features, enriching the source domain information at the data level;
[0030] S3, low-quality data generation step: performing context feature degradation and texture feature degradation processing on the reference truth data set to obtain a low-quality data set;
[0031] S4, training data generation step: performing different feature extraction processes on the texture feature attenuated data set to obtain texture feature enhancement results corresponding to different attenuation features, which are used to extract features specific to the target scene from the data;
[0032] S5, Multi-head decoding model training steps: During the training process, the feature extraction model and the multi-head decoding model are closely connected at the algorithm level by means of data enhancement to reduce the cumulative error. During the training process, the L1-norm is used as the loss function for training;
[0033] S6, step of establishing pseudo labels of attenuation features. Cluster the training data, establish the prior distribution of attenuation features, count the main attenuation features and generate pseudo labels to describe the attenuation features of the target domain observation input;
[0034] S7, observation information extraction step: establishing an attenuation feature generator to output an attenuation feature label according to the attenuation feature of the target domain;
[0035] S8, attenuation feature response step: Based on the multi-focus image fusion technology, an attenuation label responder is established to generate images and fuse significant texture features to output perceptual quality enhancement results.
[0036] Embodiment 1:
[0037] Figure 1This is a training flow chart of a method for enhancing the perception quality of underwater observation targets based on observation information guidance according to the present invention. Figure 1 As shown, the method includes:
[0038] (1) Attenuation feature establishment steps. According to the differences in attenuation characteristics of different types of water bodies, semantic labels of fuzzy attenuation, strong attenuation and multiple attenuation features are established, namely L1, L2 and L3.
[0039] In the specific implementation, the data of different pixel intensities in the training data are marked, where: i →0 is marked as L1, I i →1 blurred attenuation data is marked as L2,0 i The multiple attenuation data <1 is marked as L3, that is:
[0040] Label data{I i →0,L1}{i};
[0041] Label data{I i →1,L2}{i};
[0042] Label data{0 i <1, L3}{i}.
[0043] (2) The process of generating reference truth values of different attenuation characteristics. Figure 1 As shown in Fig. 1, different features are extracted from the reference truth data set Y as needed to obtain the reference truth values corresponding to different attenuation features, enriching the source domain information at the data level, that is, f F :Y→{Y L1 , Y L2 , Y L3}.
[0044] In the specific implementation, the original reference truth data Y is based on the feature extraction model f F Perform feature extraction to obtain the reference true values corresponding to different attenuation features, enriching the source domain information at the data level, i.e., f F :Y→{Y L1 , Y L2 , Y L3};like Figure 2 First, Y is mapped to the dual space through T1; then, the data of label L1 is normalized in the dual space. Norm In order to enhance the potential texture representation in the strongly attenuated image, the data of labels L2 and L3 are retinexed. Retinex To enhance the blurred texture details in blurred attenuation and multi-attenuation images; then, multi-scale feature fusion operation is performed simultaneouslyFusion , and perform dual inverse mapping T2 to map the data back to RGB space; then, perform Retinex operation f on the data of labels L1 and L3 in RGB space Retinex , to avoid the generation of artifact noise; finally, the output data set {Y L1 , Y L2 , Y L3}.
[0045] (3) Low-quality data generation step: The reference truth dataset is subjected to context feature degradation and texture feature degradation to obtain a low-quality dataset.
[0046] In the specific implementation, X∈{X1∪X2} is used to represent a low-quality data set, where: X1 is generated by the data set Y1 through the texture feature degenerater calculation G1, where Y1∈Y; X2 is generated by the data set Y2 through the context feature degenerater G2, where Y2∈Y.
[0047] (4) Training data generation step: The texture feature attenuated data set is processed with different feature extractions to obtain texture feature enhancement results corresponding to different attenuation features, which are used to extract features specific to the target scene.
[0048] In the specific implementation, the texture feature attenuated data set X1 is based on the feature extraction model f F Different feature extractions are performed to obtain texture feature enhancement results corresponding to different attenuation features. The processing process is as follows Figure 2 As shown, there is f F :X1→X'1={X' 1:L1 , X' 1:L2 , X' 1:L3}, x'1∈{X' 1:L1 , X' 1:L2 , X' 1:L3}.
[0049] (5) Multi-head decoding model training step. During the training process, the feature extraction model f is implemented in a data enhancement manner. F With the multi-head decoding model f D:Li (f E ) is closely connected at the algorithm level to reduce the cumulative error. During the training process, the L1 norm is used as the loss function for training.
[0050] In specific implementation, Figure 3 As shown, the dataset {X2∪X'1} is input to the multi-head decoding model f D:Li (f E ), and get the predicted output, then we have f D:Li (f E ): X'1→P1:Li , f D:Li (f E ): X2→P 2:Li .f D:Li Based on the encoding-decoding structure, a maximum pooling layer is used to perform 1 / 2 downsampling, a deconvolution layer is used to perform ×2 upsampling, and a deep cascade layer of the U-NET structure is used to implement the encoder f E With different decoders f D:Li The cross-links use the L1 norm for model optimization during training, namely:
[0051] P 1:Li =f D:Li (f E (X'1,θ E ), s E ,θ D );
[0052] P 2:Li =f D:Li (f E (X2, θ E ), s E ,θ D );
[0053] argmin Loss(Y Li , {P 1:Li , P 2:Li})=|Y Li -{P 1:Li , P 2:Li}|.
[0054] Among them, L i Used to index different attenuation feature labels {L1, L2, L3}, θ E represents the encoder parameters; θ D Indicates decoder parameters; s E represents the features transmitted across the link; Y Li The reference true value representing different attenuation characteristics is output by step (2); {P 1:Li , P 2:Li} represents the predicted output, which is output by step (5).
[0055] (6) Attenuation feature pseudo-label establishment step. Cluster the dataset {X2∪X'1} to establish the prior distribution of attenuation features. Then, count the main attenuation features and generate pseudo-labels to describe the I of the target domain observation input. in attenuation characteristics.
[0056] In specific implementation, Figure 4As shown in the figure, the dataset {X2∪X'1} (i.e., source domain data) is used to establish the prior distribution of attenuation features [ρ1, ρ2, ..., ρ n ], the k value of the clustering algorithm is set to 3; the main attenuation features are counted and pseudo labels {Label1, Label2, …, Label n}, Used to describe the target domain underwater image I in The decay characteristics of L in , where n represents the number of source domain images.
[0057] (7) Observation information extraction step: Establish an attenuation feature generator to output an attenuation feature label based on the attenuation feature of the target domain.
[0058] When performing the task of enhancing underwater images in the target domain, such as Figure 4 As shown, the attenuation tag responder G d Extract I based on K-Means clustering algorithm in The attenuation characteristic distribution of And in the pseudo-label distribution [ρ1, ρ2, ..., ρ n ] using KNN to retrieve and label and distribute The three prior distributions with the smallest deviations ρ i , and according to its significance, I in The decay characteristics of L in Label as attenuation type i , Label i ∈{L1, L2, L3}.
[0059] (8) Attenuation feature response step: Based on the multi-focus image fusion technology, an attenuation label responder is established to generate images and fuse significant texture features, outputting perceptual quality enhancement results, thereby obtaining a properly conditioned enhancement result.
[0060] In specific implementation, Figure 1 As shown in (b), firstly, the tag responder G is attenuated. r According to I in Tags in , the corresponding trained model f F 、Encoder f E 、Decoder f D:Li Output image I D:Li Marked as the fusion base, then the texture features of the remaining decoder output images are extracted based on the salient focus principle, and finally multi-focus fusion is performed to obtain the enhanced result I output .
[0061] Embodiment 2:
[0062] A method for enhancing the perception quality of underwater observation targets based on observation information, comprising:
[0063] (1) Establishment of attenuation characteristics.
[0064] (2) Generation of reference truth values for different attenuation characteristics.
[0065] (3) Low-quality data generation.
[0066] (4) Training data generation.
[0067] (5) Multi-head decoding model training.
[0068] (6) Establishment of attenuation feature pseudo-labels.
[0069] (7) Observation information extraction.
[0070] (8) Attenuation characteristic response.
[0071] Furthermore, in the step of generating the reference true values of different attenuation features, corresponding feature extraction models are designed for different attenuation labels.
[0072] Based on the duality principle, the model inverts the data set Y to highlight the degraded texture features. Normalization is used for strongly attenuated data. On the one hand, it improves I i →0 area to enhance the potential texture representation; on the other hand, different pixel intensities are mapped to the same interval to constrain the solution space of the dual function.
[0073] Furthermore, in the multi-head decoder training step, in order to improve the stability of the model, a joint optimization method is used to train the multi-head decoder. It includes:
[0074] f D The multi-output encoder-decoder f D:Li (f E ) composition, L i ∈{L1, L2, L3} is used to index different attenuation features. Figure 1 As shown in the figure, during the training process, data enhancement is used to achieve close connection at the algorithm level, and the loss function is minimized to achieve close connection at the algorithm level, thereby reducing the error accumulation caused by the cascade at the model level in the actual application process.
[0075] Furthermore, in the observation information extraction step, a semantic tag generator of attenuation features is designed, which is used according to the actual observation scene, including:
[0076] In the source domain (training set {X2∪X'1}), the prior distribution of attenuation features [ρ1, ρ2, ..., ρ n]; count its main attenuation features and generate pseudo labels {Label1, Label2, ..., Label n}, Used to describe the target domain underwater image I in The attenuation characteristics of the target domain are extracted based on the K-Means clustering algorithm. in The attenuation characteristic distribution of And in the pseudo-label distribution [ρ1, ρ2, ..., ρ n ] using KNN to retrieve and label and distribute The three prior distributions with the smallest deviations ρ i , and according to its significance, I in The decay characteristics of L in Label as attenuation type i , Label i ∈{L1, L2, L3}.
[0077] Furthermore, in the attenuation feature response step, an attenuation label responder is designed based on the multi-focus image fusion technology to guide the enhancement method to process according to the attenuation features of the observed object in the actual scene, including:
[0078] Marking context features, constructing salient texture feature maps, and multi-focus fusion. Attenuated label responder G r According to I in Tags in , the corresponding trained decoder f D:Li Output image I D:Li It is marked as the fusion base, and then the texture features of the remaining decoder output images are extracted based on the salient focus principle, and finally multi-focus fusion is performed to obtain the enhanced result.
[0079] The underwater target perception enhancement method based on the guidance of target domain observation information proposed in the present invention can solve the ill-posed enhancement problem caused by the quality difference between the target domain and the source domain. The present invention uses the attenuation characteristics of the target domain image as semantic labels to guide the source domain model to generate and fuse significant texture features and context features, thereby achieving the well-posed enhancement of underwater images under different attenuation conditions, which has important application value in the fields of underwater target detection, key object perception, autonomous perception robots, etc.
[0080] The method proposed in the present invention takes the attenuation characteristics of the target domain image as semantic labels, which can guide the source domain enhancement model to generate images and fuse significant texture features, thereby achieving adaptive enhancement of the perception quality of underwater observation targets under different attenuation conditions, improving the operational performance of the visual system, and having important application value in the fields of underwater target detection, key object perception, and autonomous perception robots.
[0081] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A method for enhancing underwater target perception based on target domain observation information guidance. It is characterized in that The method comprises the following steps: S1. Attenuation feature establishment step: according to the attenuation feature differences of different types of water bodies, semantic labels of fuzzy attenuation, strong attenuation and multiple attenuation features are established; S2, reference truth value generation step of different attenuation features: extract different features from the reference truth value data set to obtain reference truth values corresponding to different attenuation features, enriching the source domain information at the data level; S3, low-quality data generation step: performing context feature degradation and texture feature degradation processing on the reference truth data set to obtain a low-quality data set; S4, training data generation step: performing different feature extraction processes on the texture feature attenuated data set to obtain texture feature enhancement results corresponding to different attenuation features, which are used to extract features specific to the target scene from the data; S5. Multi-head decoding model training steps: During the training process, the feature extraction model and the multi-head decoding model are closely connected at the algorithm level by means of data enhancement. During the training process, L 1 -norm is used as the loss function for training; S6, attenuation feature pseudo-label establishment step: clustering the training data, establishing a priori distribution of the attenuation feature, counting its main attenuation features and generating pseudo-labels to describe the attenuation features of the target domain observation input; S7, observation information extraction step: establishing an attenuation feature generator to output an attenuation feature label according to the attenuation feature of the target domain; S8, attenuation feature response step: Based on the multi-focus image fusion technology, an attenuation label responder is established to generate images and fuse significant texture features to output perceptual quality enhancement results.
2. The underwater target perception enhancement method based on target domain observation information guidance as claimed in claim 1, It is characterized in that S1 includes: marking the data of different pixel intensities in the training data, wherein: i →0 strong attenuation data is marked as L 1 , I i →The blurred attenuation data of 1 is marked as L 2 , 0 i Multiple decay data < 1 are marked as L 3 ,Right now: Label data{I i →0,L 1 }{i}; Label data{I i →1,L 2 }{i}; Label data{0<I i <1,L 3 }{i}。 3. The underwater target perception enhancement method based on target domain observation information guidance as claimed in claim 1, It is characterized in that The S2 includes: extracting the original reference truth data Y based on the feature extraction model f F Perform feature extraction to obtain the reference true values corresponding to different attenuation features, enriching the source domain information at the data level, i.e., f F :Y→{Y L1 , Y L2 , Y L3 }: First, pass Y through T 1 Map to the dual space; then, in the dual space, label L 1 Normalize the data in Layer Norm To enhance the potential texture representation in strongly attenuated images, the label L 2 , L 3 Retinex operation on the data Retinex To enhance the blurred texture details in blurred attenuation and multi-attenuation images; then, multi-scale feature fusion operation is performed simultaneously Fusion , and perform the dual inverse mapping T 2 , map the data back to RGB space; then, 1 , L 3 The data is then retinexed in RGB space. Retinex , to avoid the generation of artifact noise; finally, the output data set {Y L1 , Y L2 , Y L3 }.
4. The underwater target perception enhancement method based on target domain observation information guidance as claimed in claim 3, It is characterized in that The S3 includes: using X∈{X 1 ∪X 2 } represents a low-quality dataset, where: X 1 From the dataset Y 1 After the texture feature degrader, G is calculated 1 Generate, where Y 1 ∈Y;X 2 From the dataset Y 2 After the context feature degenerater G 2 Generate, where Y 2 ∈Y.
5. The underwater target perception enhancement method based on target domain observation information guidance as claimed in claim 4, It is characterized in that The S4 comprises: attenuating the texture feature data set X 1 Based on the feature extraction model f F Perform different feature extraction to obtain texture feature enhancement results corresponding to different attenuation features; that is, f F :X 1 →X' 1 ={X' 1:L1 , X' 1:L2 , X' 1:L3 }、x' 1 ∈{X' 1:L1 , X' 1:L2 , X' 1:L3 }.
6. The underwater target perception enhancement method based on target domain observation information guidance as claimed in claim 5, It is characterized in that The S5 comprises: transforming the data set {X 2 ∪X' 1 } Input to the multi-head decoding model f D:Li (f E ), and get the predicted output, then we have f D:Li (f E ): X' 1 →P 1:Li , f D:Li (f E ):X 2 →P 2:Li ; f D:Li Based on the encoding-decoding structure, a maximum pooling layer is used to perform 1 / 2 downsampling, a deconvolution layer is used to perform ×2 upsampling, and a deep cascade layer of the U-NET structure is used to implement the encoder f E With different decoders f D:Li The cross-link is used during training. 1 The model is optimized using the norm, namely: P 1:Li =f D:Li (f E (X' 1 ,the E ),s E ,the D ); P 2:Li =f D:Li (f E (X 2 ,the E ),s E ,the D ); argmin Loss(Y Li ,{P 1:Li ,P 2:Li })=|Y Li -{P 1:Li ,P 2:Li }| Among them, L i Used to index different attenuation feature labels {L 1 , L 2 , L 3 }, θ E represents the encoder parameters; θ D Indicates decoder parameters; s E represents the features transmitted across the link; Y Li The reference true values representing different attenuation characteristics are output by step S2; {P 1:Li , P 2:Li } represents the predicted output, which is output by step S5.
7. The underwater target perception enhancement method based on target domain observation information guidance according to claim 6, It is characterized in that S6 includes: converting the source domain dataset {X 2 ∪X' 1 }Establishing the prior distribution of attenuation features based on clustering algorithm [ρ 1 , ρ 2 , ..., ρ n ], the k value of the clustering algorithm is set to 3; the main attenuation features are counted and pseudo labels {Label 1 、Label 2 、…、Label n }, Used to describe the target domain underwater image I in The decay characteristics of L in , where n represents the number of source domain images.
8. The underwater target perception enhancement method based on target domain observation information guidance as claimed in claim 7, It is characterized in that The S7 includes: when performing the target domain underwater image enhancement task, attenuating the tag responder G d Extraction based on clustering algorithm in The attenuation characteristic distribution of And in the pseudo label distribution [ρ 1 , ρ 2 , ..., ρ n ] using KNN to retrieve and label and distribute The three prior distributions with the smallest deviations ρ i , and according to its significance, I in The decay characteristics of L in Label as attenuation type i , Label i ∈{L 1 , L 2 , L 3 }.
9. The underwater target perception enhancement method based on target domain observation information guidance as claimed in claim 8, It is characterized in that The clustering algorithm is K-Means clustering algorithm.
10. The underwater target perception enhancement method based on target domain observation information guidance according to claim 8, It is characterized in that The step S8 includes: firstly attenuating the tag responder G r According to I in Tags in , the corresponding trained model f F 、Encoder f E 、Decoder f D:Li Output image I D:Li Marked as the fusion base, then the texture features of the remaining decoder output images are extracted based on the salient focus principle, and finally multi-focus fusion is performed to obtain the enhanced result I output .
Citation Information
Patent Citations
Underwater camouflage object image segmentation method based on double-branch decoder network
CN116721112A
Two-stage underwater image enhancement method based on target detection perception feature fusion
CN117974509A
Methods and Apparatus for Autonomous Robotic Control
US20170024877A1
