CN111914948B - Ocean current machine blade attachment self-adaptive identification method based on rough and fine semantic segmentation network - Google Patents
Ocean current machine blade attachment self-adaptive identification method based on rough and fine semantic segmentation network Download PDFInfo
- Publication number
- CN111914948B CN111914948B CN202010842698.7A CN202010842698A CN111914948B CN 111914948 B CN111914948 B CN 111914948B CN 202010842698 A CN202010842698 A CN 202010842698A CN 111914948 B CN111914948 B CN 111914948B
- Authority
- CN
- China
- Prior art keywords
- fine
- segmentation
- rough
- network
- training
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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
- G06V10/267—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 by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Multimedia (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the field of marine engine fault identification, and particularly relates to a marine engine blade attachment self-adaptive identification method based on a rough and fine semantic segmentation network, which comprises the following implementation steps: acquiring a sea current machine image dataset; expanding an original data set and carrying out normalization pretreatment; distributing training, verifying and testing data sets in proportion; setting up a rough and fine semantic segmentation network fused with a rough segmentation branch and a fine segmentation branch, and setting the initial fusion weight ratio to be 1:1; the rough segmentation fusion weight is kept unchanged, the fine segmentation fusion weight is added into a cross entropy loss function in an L2 regularized form, and self-adaptive descent is carried out along with the loss function; training a rough and fine semantic segmentation network; and testing the trained network to finish the fine identification of the attachments of the ocean current machine blades. The invention can accurately identify clear sea current machine images, and can refine contours of sea current machine images with larger motion blur so as to improve the identification accuracy.
Description
Technical Field
The invention relates to the field of marine engine fault identification, in particular to a marine engine blade attachment self-adaptive identification method based on a rough and fine semantic segmentation network.
Background
With the continuous reduction of the storage amount of traditional fossil energy such as coal, petroleum, natural gas and the like, more and more countries in the world begin to pay attention to the development of new energy technologies. In recent years, ocean current energy is regarded as a green and clean energy source which is buried in the ocean and is paid attention to by researchers due to the advantages of high energy density, predictability and the like, and ocean current machines are power generation devices which convert ocean current energy into electric energy. As the marine microorganism is soaked in the seawater for a long time, the marine microorganism is easy to grow on the surface of the blade of the ocean current machine to form attachments. When the attachments accumulate to a certain degree, the unbalanced faults of the blades are likely to be caused, and the power generation quality is seriously affected. Therefore, research on a method for identifying attachments of the blades of the ocean current machine with high accuracy is beneficial to maintenance personnel to timely clean the blades so as to prevent the faults.
Until now, there are relatively few methods for identifying attachments to blades of ocean current machines, which are mainly divided into two main types, one is a method based on image identification, and the other is a method based on image semantic segmentation. The method based on image recognition mainly carries out classification recognition on the attachment degree of the ocean current machine blade; compared with image recognition, the image semantic segmentation-based method can more intuitively segment each target area (such as background, blades and attachments) in the image. However, the current image semantic segmentation-based method has low accuracy in identifying the sea current machine image with motion blur acquired at high flow rate.
Disclosure of Invention
The invention provides a self-adaptive identification method for the attachments of the blades of the ocean current machine based on a rough and fine semantic segmentation network, which aims to solve the problems of the identification method for the attachments of the blades of the ocean current machine based on the semantic segmentation of images and realize more accurate identification of the attachments.
The ocean current machine blade attachment self-adaptive identification method based on the rough and fine semantic segmentation network comprises the following steps of:
step 1: the method comprises the steps of respectively collecting the underwater images of the ocean current machines with different attachment degrees under the conditions of low flow rate and high flow rate, then carrying out pixel-level semantic labeling on the collected images, and respectively labeling the background, the blades and the attachment areas in each image by using pixels 0,1 and 2, so that an image-label pair data set can be obtained.
Step 2: the image feature richness is improved by adopting two data expansion strategies of horizontal overturning and rotation enhancement, and in order to accelerate the convergence speed of subsequent network training, all images in an expansion data set are subjected to the following normalization pretreatment:
Wherein x represents a pixel matrix of a color (RGB) ocean current machine image; x max,xmin represents the maximum and minimum pixel values in the pixel matrix, respectively; x will be normalized to x normalized, which range from [ -1,1].
The preprocessed image-label pair dataset is then divided into training, validation, test datasets in a ratio of 5:1:1.
Step 3: building a coarse-fine semantic segmentation network, wherein the network merges two parts: coarse-split branches and fine-split branches; the rough segmentation branches mainly complete global identification of the ocean current machine image, and rough background, blades and attachment areas are segmented; the fine segmentation branch is responsible for extracting the image edge contour in the initial stage of network training, and is mainly responsible for carrying out contour refinement on a rough result graph generated by the rough segmentation branch in the later stage of network training so as to reduce the false recognition rate.
Specifically, the rough segmentation branch adopts SegNet networks with coding-decoding symmetrical structures, which comprise coding layers 1-5 and corresponding decoding layers 1-5, all convolution kernels are 3*3, and the class boundary information is reserved through a maximum pooling index value storage technology; to further speed up network convergence, each convolutional layer adds a batch normalization operation, and initializes the code layer portion (the first 13-layer convolution of VGG 16) using ImageNet pre-training weights; a 50% dropout layer is inserted between the coding layer and the decoding layer to prevent the network from training a fit; to achieve optimal fusion of the coarse and fine split branches, the softmax activation function convolved by the last 3*3 of the decoding layers is replaced with the ReLU activation function.
The fine division branch is an asymmetric coding-decoding structure, the coding layer directly uses the first two coding layers of the coarse division branch, the decoding layer carries out up-sampling on the output characteristics of the two coding layers and then cascades the output characteristics together, and then a 3*3 convolution layer 1 with a ReLU activation function is connected in parallel.
After weighted fusion of the result graphs of the coarse and fine branches, the final semantic segmentation graph is output using convolution layer 2 with softmax activation function.
Step 4: in order to ensure that the global identification of the ocean current machine image and the extraction of the edge contour have the same importance in the initial stage of network training, setting the initial proportion of fusion weights of coarse segmentation and fine segmentation to be 1:1; in order to enable the fine segmentation to play a role in contour refining at the later stage of network training, an L2 regularization term related to the fine segmentation fusion weight is added into the integral loss function to adaptively reduce the edge contour extraction effect of the fine segmentation branches, so that the refining effect on the image contour is enhanced.
Specifically, the overall loss function comprises two parts, namely cross entropy loss and an L2 regularization term related to refined segmentation fusion weight; thus, the overall loss function is defined as follows:
Losstotal＝Losscross_entropy+λβ2
Where Loss total represents the overall Loss function; loss cross_entropy represents the cross entropy Loss function; λβ 2 represents the L2 regularization term for the refined segmentation fusion weights 0 < β.ltoreq.1.0, and λ > 0 represents the regularization coefficient.
Specifically, the cross entropy Loss function Loss cross_entropy is defined as follows:
Wherein M is the number of target types to be marked in the ocean current machine image, and M is 3 according to the knowledge of the step 1; when the semantic tag c is determined to be one of {0,1,2}, y c is taken as 1; p c represents the softmax probability predicted as category c; log (·) represents a logarithmic function.
Specifically, the softmax probability p c is defined as follows:
wherein, o r and o c respectively represent the eigenvalues of the category r and the category c corresponding to the eigenvector o extracted by the convolution of the last layer in the rough and fine semantic segmentation network; exp (·) represents an exponential function.
To prevent β from updating too much or too little during network training, the updated value of β each time is limited to the range of [0.3,1] by the following constraints:
Wherein |β| represents the absolute value of β;0.3 and 1.0 respectively represent an optimization target value and an upper limit value of a preset fine segmentation fusion weight; epsilon > 0 is a positive number close to 0 for preventing numerical problems caused by 0 denominator; beta constrained represents the constrained value of beta; before the constraint is performed, the updated value of β needs to be taken as an absolute value to ensure that β is a non-negative number.
Step 5: inputting a training data set and a verification data set into a built rough and fine semantic segmentation network, setting initial values of fusion weights of rough and fine segmentation to be 1.0, fixing the fusion weights of the rough segmentation, and presetting a fusion weight optimization target value of the fine segmentation to be 0.3; the network is adaptively trained using Adadelta optimizers.
Step 6: after the training iteration number reaches a set maximum value, ending the network training; loading the trained network model weight and the fused weight of the fine segmentation into a coarse and fine semantic segmentation network, then using a test data set to perform network test, and outputting a recognition result graph of the marine engine blade attachments.
Advantageous effects
Compared with the prior art, the invention has the following technical effects:
1) The invention adopts two data expansion strategies of horizontal overturning and rotation enhancement to simulate the motion process of the ocean current machine, improves the feature richness of the image and is beneficial to the network learning of more useful information.
2) Compared with other single-path semantic segmentation networks, the coarse-fine semantic segmentation network provided by the invention has two paths of coarse and fine segmentation branches.
3) The invention adopts the rough and fine segmentation branches with coding-decoding structures, wherein the rough segmentation branch adopts a symmetrical SegNet network architecture, and the class boundary information is reserved by a maximum pooling index value storage technology; the precise segmentation branch adopts asymmetric up-sampling and feature cascading technology to realize the identification of the image edge contour.
4) The self-adaptive recognition method provided by the invention can automatically enhance the contour refining capability of the fine segmentation branches in the network training process; after training, the fine segmentation branch has the capability of refining the edge contour of the rough result graph output by the coarse segmentation branch.
5) The invention can accurately identify the clear ocean current machine image acquired at low rotation speed and can accurately identify the ocean current machine image with motion blur acquired at high rotation speed.
Drawings
The foregoing and/or additional aspects and advantages of the invention will become apparent and may be better understood from the following description of embodiments taken in conjunction with the accompanying drawings in which:
FIG. 1 is a schematic algorithm flow diagram of a self-adaptive identification method for attachment of a marine engine blade based on a rough-fine semantic segmentation network;
FIG. 2 illustrates a portion of a current machine image sample acquired at low and high flow rates;
Fig. 3 is a schematic diagram of architecture of a coarse-fine semantic segmentation network proposed in the present invention.
Detailed Description
The following detailed description of embodiments of the invention is exemplary and intended to be illustrative of the invention and not to be construed as limiting the invention.
As shown in FIG. 1, the invention provides a marine engine blade attachment self-adaptive identification method based on a rough and fine semantic segmentation network, which comprises the following steps:
Step 1: respectively acquiring underwater images of the ocean current machines with different attachment degrees under the conditions of low flow rate and high flow rate, as shown in figure 2; and then, carrying out pixel-level semantic labeling on the acquired images, wherein the background, the blades and the attachment areas in each image are respectively labeled by pixels 0, 1 and 2, so that an image-label pair data set can be obtained.
Step 2: the image feature richness is improved by adopting two data expansion strategies of horizontal overturning and rotation enhancement, and in order to accelerate the convergence speed of subsequent network training, all images in an expansion data set are subjected to the following normalization pretreatment:
Wherein x represents a pixel matrix of a color (RGB) ocean current machine image; x max,xmin represents the maximum and minimum pixel values in the pixel matrix, respectively; x will be normalized to x normalized, which range from [ -1,1].
The preprocessed image-label pair dataset is then divided into training, validation, test datasets in a ratio of 5:1:1.
Step 3: building a coarse-fine semantic segmentation network, wherein the network merges two parts: coarse-split branches and fine-split branches; the rough segmentation branches mainly complete global identification of the ocean current machine image, and rough background, blades and attachment areas are segmented; the fine segmentation branch is responsible for extracting the image edge contour in the initial stage of network training, and is mainly responsible for carrying out contour refinement on a rough result graph generated by the rough segmentation branch in the later stage of network training so as to reduce the false recognition rate.
In this embodiment, the rough and fine semantic segmentation network merges two parts, namely a rough segmentation branch and a fine segmentation branch, as shown in fig. 3; specifically, the rough segmentation branch adopts SegNet networks with coding-decoding symmetrical structures, which comprise coding layers 1-5 and corresponding decoding layers 1-5, all convolution kernels are 3*3, and the class boundary information is reserved through a maximum pooling index value storage technology; to further speed up network convergence, each convolutional layer adds a batch normalization operation, and initializes the code layer portion (the first 13-layer convolution of VGG 16) using ImageNet pre-training weights; a 50% dropout layer is inserted between the coding layer and the decoding layer to prevent the network from training a fit; to achieve optimal fusion of the coarse and fine split branches, the softmax activation function convolved by the last 3*3 of the decoding layers is replaced with the ReLU activation function.
The fine division branch is an asymmetric coding-decoding structure, the coding layer directly uses the first two coding layers of the coarse division branch, the decoding layer carries out up-sampling on the output characteristics of the two coding layers and then cascades the output characteristics together, and then a 3*3 convolution layer 1 with a ReLU activation function is connected in parallel.
After weighted fusion of the result graphs of the coarse and fine branches, the final semantic segmentation graph is output using convolution layer 2 with softmax activation function.
Step 4: in order to ensure that the global identification of the ocean current machine image and the extraction of the edge contour have the same importance in the initial stage of network training, setting the initial proportion of fusion weights of coarse segmentation and fine segmentation to be 1:1; in order to enable the fine segmentation to play a role in contour refining at the later stage of network training, an L2 regularization term related to the fine segmentation fusion weight is added into the integral loss function to adaptively reduce the edge contour extraction effect of the fine segmentation branches, so that the refining effect on the image contour is enhanced.
In this embodiment, the overall loss function includes two parts, namely, cross entropy loss and L2 regularization term about refined segmentation fusion weight; thus, the overall loss function is defined as follows:
Losstotal＝Losscross_entropy+λβ2
Where Loss total represents the overall Loss function; loss cross_entropy represents the cross entropy Loss function; λβ 2 represents the L2 regularization term for the refined segmentation fusion weights 0 < β.ltoreq.1.0, and λ > 0 represents the regularization coefficient.
Specifically, the cross entropy Loss function Loss cross_entropy is defined as follows:
Wherein M is the number of target types to be marked in the ocean current machine image, and M is 3 according to the knowledge of the step 1; when the semantic tag c is determined to be one of {0,1,2}, y c is taken as 1; p c represents the softmax probability predicted as category c; log (·) represents a logarithmic function.
Specifically, the softmax probability p c is defined as follows:
wherein, o r and o c respectively represent the eigenvalues of the category r and the category c corresponding to the eigenvector o extracted by the convolution of the last layer in the rough and fine semantic segmentation network; exp (·) represents an exponential function.
To prevent β from updating too much or too little during network training, the updated value of β each time is limited to the range of [0.3,1] by the following constraints:
Wherein |β| represents the absolute value of β;0.3 and 1.0 respectively represent an optimization target value and an upper limit value of a preset fine segmentation fusion weight; epsilon > 0 is a positive number close to 0 for preventing numerical problems caused by 0 denominator; beta constrained represents the constrained value of beta; before the constraint is performed, the updated value of β needs to be taken as an absolute value to ensure that β is a non-negative number.
Step 5: inputting a training data set and a verification data set into a built rough and fine semantic segmentation network, setting initial values of fusion weights of rough and fine segmentation to be 1.0, fixing the fusion weights of the rough segmentation, and presetting a fusion weight optimization target value of the fine segmentation to be 0.3; the network is adaptively trained using Adadelta optimizers.
Step 6: after the training iteration number reaches a set maximum value, ending the network training; loading the trained network model weight and the fused weight of the fine segmentation into a coarse and fine semantic segmentation network, then using a test data set to perform network test, and outputting a recognition result graph of the marine engine blade attachments.
Claims (2)
1. A marine engine blade attachment self-adaptive identification method based on a rough and fine semantic segmentation network is characterized by comprising the following steps:
Step 1: collecting underwater images of the ocean current machines with different attachment degrees under the conditions of low flow rate and high flow rate respectively, then carrying out pixel-level semantic labeling on the collected images, and labeling the background, the blades and the attachment areas in each image by using pixels 0,1 and 2 respectively, so that an image-tag pair data set can be obtained;
Step 2: the image feature richness is improved by adopting two data expansion strategies of horizontal overturning and rotation enhancement, and in order to accelerate the convergence speed of subsequent network training, all images in an expansion data set are subjected to the following normalization pretreatment:
Wherein x represents a pixel matrix of a color (RGB) ocean current machine image; x max,xmin represents the maximum and minimum pixel values in the pixel matrix, respectively; x is normalized to x normalized, which is in the range of [ -1,1];
dividing the preprocessed image-label pair data set into training, verifying and testing data sets according to the proportion of 5:1:1;
Step 3: building a coarse-fine semantic segmentation network, wherein the network merges two parts: coarse-split branches and fine-split branches; the rough segmentation branches mainly complete global identification of the ocean current machine image, and rough background, blades and attachment areas are segmented; the fine segmentation branch is responsible for extracting the edge contour of the image in the initial stage of network training, and is mainly responsible for carrying out contour refining on a rough result graph generated by the rough segmentation branch in the later stage of network training so as to reduce the false recognition rate;
Specifically, the rough segmentation branch adopts SegNet networks with coding-decoding symmetrical structures, all convolution kernels are 3*3, and the class boundary information is reserved through a maximum pooling index value storage technology; to further speed up network convergence, each convolutional layer adds a batch normalization operation, and initializes the code layer portion, the first 13 layer convolutions of VGG16, using ImageNet pre-training weights; a 50% dropout layer is inserted between the coding layer and the decoding layer to prevent the network from training a fit; in order to realize the optimal fusion of the coarse segmentation branch and the fine segmentation branch, the softmax activation function convolved by the last 3*3 of the decoding layer is replaced by the ReLU activation function;
The fine segmentation branch is an asymmetric coding-decoding structure, the coding layer directly uses the first two coding layers of the coarse segmentation branch, the decoding layer carries out up-sampling on the output characteristics of the two coding layers and then cascades the output characteristics together, and then a 3*3 convolution layer with a ReLU activation function is connected in parallel;
After the result graphs of the coarse branches and the fine branches are subjected to weighted fusion, a final semantic segmentation graph is output by using a convolution layer with a softmax activation function;
Step 4: in order to ensure that the global identification of the ocean current machine image and the extraction of the edge contour have the same importance in the initial stage of network training, setting the initial proportion of fusion weights of coarse segmentation and fine segmentation to be 1:1; in order to enable the fine segmentation to exert the contour refining effect in the later stage of network training, adding an L2 regularization term related to the fine segmentation fusion weight into the integral loss function to adaptively reduce the edge contour extraction effect of the fine segmentation branches, so that the refining effect on the image contour is enhanced;
Step 5: inputting a training data set and a verification data set into a built rough and fine semantic segmentation network, setting initial values of fusion weights of rough and fine segmentation to be 1.0, fixing the fusion weights of the rough segmentation, and presetting a fusion weight optimization target value of the fine segmentation to be 0.3; adaptively training the network by using Adadelta optimizers;
step 6: after the training iteration number reaches a set maximum value, ending the network training; loading the trained network model weight and the fused weight of the fine segmentation into a coarse and fine semantic segmentation network, then using a test data set to perform network test, and outputting a recognition result graph of the marine engine blade attachments.
2. The adaptive identification method of marine engine blade attachments based on a rough and fine semantic segmentation network according to claim 1, wherein the overall loss function in step4 comprises two parts, namely cross entropy loss and an L2 regularization term about fine segmentation fusion weight; thus, the overall loss function is defined as follows:
Losstotal＝Losscross_entropy+λβ2
Where Loss total represents the overall Loss function; loss cross_entropy represents the cross entropy Loss function; λβ 2 represents an L2 regularization term with respect to the refined segmentation fusion weight 0 < β.ltoreq.1.0, and λ > 0 represents a regularization coefficient;
Specifically, the cross entropy Loss function Loss cross_entropy is defined as follows:
wherein M is the number of target types to be marked in the ocean current machine image, and M is 3 according to the knowledge of the step 1; when the semantic tag c is determined to be one of {0,1,2}, y c is taken as 1; p c represents the softmax probability predicted as category c; log (·) represents a logarithmic function;
specifically, the softmax probability p c is defined as follows:
Wherein, o r and o c respectively represent the eigenvalues of the category r and the category c corresponding to the eigenvector o extracted by the convolution of the last layer in the rough and fine semantic segmentation network; exp (·) denotes an exponential function;
to prevent β from updating too much or too little during network training, the updated value of β each time is limited to the range of [0.3,1] by the following constraints:
Wherein |β| represents the absolute value of β;0.3 and 1.0 respectively represent an optimization target value and an upper limit value of a preset fine segmentation fusion weight; epsilon > 0 is a positive number close to 0 for preventing numerical problems caused by 0 denominator; beta constrained represents the constrained value of beta; before the constraint is performed, the updated value of β needs to be taken as an absolute value to ensure that β is a non-negative number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010842698.7A CN111914948B (en) | 2020-08-20 | 2020-08-20 | Ocean current machine blade attachment self-adaptive identification method based on rough and fine semantic segmentation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010842698.7A CN111914948B (en) | 2020-08-20 | 2020-08-20 | Ocean current machine blade attachment self-adaptive identification method based on rough and fine semantic segmentation network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111914948A CN111914948A (en) | 2020-11-10 |
| CN111914948B true CN111914948B (en) | 2024-07-26 |
Family
ID=73279152
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010842698.7A Active CN111914948B (en) | 2020-08-20 | 2020-08-20 | Ocean current machine blade attachment self-adaptive identification method based on rough and fine semantic segmentation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111914948B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112950617B (en) * | 2021-03-24 | 2024-05-10 | 上海海事大学 | Tidal current machine blade attachment identification method based on continuous rotation image enhancement and condition generation countermeasure network |
| CN113989549B (en) * | 2021-10-21 | 2025-03-07 | 神思电子技术股份有限公司 | A pseudo-label-based semi-supervised learning image classification optimization method and system |
| CN114511576B (en) * | 2022-04-19 | 2022-07-12 | 山东建筑大学 | Scale-adaptive feature-enhanced deep neural network for image segmentation method and system |
| CN115810020B (en) * | 2022-12-02 | 2023-06-02 | 中国科学院空间应用工程与技术中心 | A coarse-to-fine remote sensing image segmentation method and system based on semantic guidance |
| CN116152493A (en) * | 2023-02-14 | 2023-05-23 | 南京邮电大学 | A method for identifying surface types to be cleaned based on semantic segmentation and cleaning equipment for implementing the method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110863935A (en) * | 2019-11-19 | 2020-03-06 | 上海海事大学 | Recognition method of blade attachments of ocean current machine based on VGG16-SegUnet and dropout |
| CN111462126A (en) * | 2020-04-08 | 2020-07-28 | 武汉大学 | Semantic image segmentation method and system based on edge enhancement |
| CN112132777A (en) * | 2020-08-11 | 2020-12-25 | 浙江工业大学 | A Multiresolution Perceptual Learning-Based Approach for Central Venous Catheter Segmentation and Tip Localization |
| CN112950617A (en) * | 2021-03-24 | 2021-06-11 | 上海海事大学 | Tidal current machine blade attachment identification method based on continuous rotation image enhancement and conditional generation countermeasure network |
| CN114998307A (en) * | 2022-07-06 | 2022-09-02 | 重庆大学 | Two-stage full 3D abdominal organ segmentation method and system based on dual-resolution network |
| CN115719475A (en) * | 2022-10-24 | 2023-02-28 | 北京交通大学 | Three-stage trackside equipment fault automatic detection method based on deep learning |
Family Cites Families (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2002951574A0 (en) * | 2002-09-20 | 2002-10-03 | Unisearch Limited | Method of signalling motion information for efficient scalable video compression |
| US10019657B2 (en) * | 2015-05-28 | 2018-07-10 | Adobe Systems Incorporated | Joint depth estimation and semantic segmentation from a single image |
| US9741125B2 (en) * | 2015-10-28 | 2017-08-22 | Intel Corporation | Method and system of background-foreground segmentation for image processing |
| CN105513041B (en) * | 2015-10-28 | 2018-12-21 | 深圳大学 | A kind of method and system of large format remote sensing images sea land segmentation |
| CN110838124B (en) * | 2017-09-12 | 2021-06-18 | 深圳科亚医疗科技有限公司 | Method, system, and medium for segmenting images of objects having sparse distribution |
| WO2019136623A1 (en) * | 2018-01-10 | 2019-07-18 | Nokia Technologies Oy | Apparatus and method for semantic segmentation with convolutional neural network |
| CN108062756B (en) * | 2018-01-29 | 2020-04-14 | 重庆理工大学 | Image Semantic Segmentation Based on Deep Fully Convolutional Networks and Conditional Random Fields |
| CN108805889B (en) * | 2018-05-07 | 2021-01-08 | 中国科学院自动化研究所 | Edge-guided segmentation method, system and equipment for refined salient objects |
| CN108898140A (en) * | 2018-06-08 | 2018-11-27 | 天津大学 | Brain tumor image segmentation algorithm based on improved full convolutional neural networks |
| US10304193B1 (en) * | 2018-08-17 | 2019-05-28 | 12 Sigma Technologies | Image segmentation and object detection using fully convolutional neural network |
| CN109271895B (en) * | 2018-08-31 | 2022-03-04 | 西安电子科技大学 | Pedestrian re-identification method based on multi-scale feature learning and feature segmentation |
| CN110880001A (en) * | 2018-09-06 | 2020-03-13 | 银河水滴科技（北京）有限公司 | Training method, device and storage medium for semantic segmentation neural network |
| CN109509192B (en) * | 2018-10-18 | 2023-05-30 | 天津大学 | Semantic segmentation network integrating multi-scale feature space and semantic space |
| CN109711413B (en) * | 2018-12-30 | 2023-04-07 | 陕西师范大学 | Image semantic segmentation method based on deep learning |
| CN110033021B (en) * | 2019-03-07 | 2021-04-06 | 华中科技大学 | Fault classification method based on one-dimensional multipath convolutional neural network |
| CN111091130A (en) * | 2019-12-13 | 2020-05-01 | 南京邮电大学 | Real-time image semantic segmentation method and system based on lightweight convolutional neural network |
| CN111127499B (en) * | 2019-12-20 | 2024-08-23 | 北京工业大学 | Safety inspection image cutter detection and segmentation method based on semantic contour information |
| CN111080688A (en) * | 2019-12-25 | 2020-04-28 | 左一帆 | Depth map enhancement method based on depth convolution neural network |
| CN111340828A (en) * | 2020-01-10 | 2020-06-26 | 南京航空航天大学 | Brain Glioma Segmentation Based on Cascaded Convolutional Neural Networks |
-
2020
- 2020-08-20 CN CN202010842698.7A patent/CN111914948B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110863935A (en) * | 2019-11-19 | 2020-03-06 | 上海海事大学 | Recognition method of blade attachments of ocean current machine based on VGG16-SegUnet and dropout |
| CN111462126A (en) * | 2020-04-08 | 2020-07-28 | 武汉大学 | Semantic image segmentation method and system based on edge enhancement |
| CN112132777A (en) * | 2020-08-11 | 2020-12-25 | 浙江工业大学 | A Multiresolution Perceptual Learning-Based Approach for Central Venous Catheter Segmentation and Tip Localization |
| CN112950617A (en) * | 2021-03-24 | 2021-06-11 | 上海海事大学 | Tidal current machine blade attachment identification method based on continuous rotation image enhancement and conditional generation countermeasure network |
| CN114998307A (en) * | 2022-07-06 | 2022-09-02 | 重庆大学 | Two-stage full 3D abdominal organ segmentation method and system based on dual-resolution network |
| CN115719475A (en) * | 2022-10-24 | 2023-02-28 | 北京交通大学 | Three-stage trackside equipment fault automatic detection method based on deep learning |
Non-Patent Citations (4)
| Title |
|---|
| An adaptive coarse-fine semantic segmentation method for the attachment recognition on marine current turbines;Peng Haiyang等;COMPUTERS & ELECTRICAL ENGINEERING;20210610;第93卷;全文 * |
| SSFNET-VOS: Semantic segmentation and fusion network for video object segmentation;Sharma, Vipal Kumar等;PATTERN RECOGNITION LETTERS;20201231;第140卷;全文 * |
| 基于MobileNet的海流透平机叶片附着物图像诊断;刘然等;控制工程;20210325;第29卷(第10期);全文 * |
| 基于语义分割网络的潮流机叶片附着物识别方法研究;彭海洋;中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑;20220515(第5期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111914948A (en) | 2020-11-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111914948B (en) | Ocean current machine blade attachment self-adaptive identification method based on rough and fine semantic segmentation network | |
| CN111680706B (en) | A Two-channel Output Contour Detection Method Based on Encoding and Decoding Structure | |
| CN108229550B (en) | Cloud picture classification method based on multi-granularity cascade forest network | |
| CN115272776B (en) | Hyperspectral image classification method based on double-path convolution and double attention and storage medium | |
| CN111209907B (en) | Artificial intelligent identification method for product characteristic image in complex light pollution environment | |
| CN114758133B (en) | Image defect segmentation method based on superpixel active learning and semi-supervised learning strategy | |
| CN114092742B (en) | Multi-angle-based small sample image classification device and method | |
| CN114066867B (en) | Segmentation method of crack extension trace missing region based on deep learning | |
| CN110863935B (en) | Recognition method of blade attachments of ocean current machine based on VGG16-SegUnet and dropout | |
| CN116071352A (en) | A method for generating surface defect images of electrical safety appliances | |
| CN116912708A (en) | Remote sensing image building extraction method based on deep learning | |
| CN108510001A (en) | A kind of blade of wind-driven generator defect classification method and its categorizing system | |
| CN115760619A (en) | A method and system for quickly removing highlights on smooth surfaces by combining U2-Net and LaMa models | |
| CN118506263A (en) | A foreign body detection method for power transmission lines in complex environments based on deep learning | |
| CN119559435A (en) | Multi-modal small sample image classification method and system based on multi-scale dynamic feature fusion | |
| CN115376066B (en) | Airport scene target detection multi-weather data enhancement method based on improvement CycleGAN | |
| CN110189330A (en) | A method of background removal based on deep learning | |
| An | Xception network for weather image recognition based on transfer learning | |
| CN120301739A (en) | A signal modulation recognition method and device based on deep learning | |
| CN112487933A (en) | Radar waveform identification method and system based on automatic deep learning | |
| CN119850633A (en) | Industrial defect detection model training method, detection method, robot and storage medium | |
| CN118334159A (en) | Electric power image generation model construction method, image generation method and related device | |
| CN113610159A (en) | Small sample photovoltaic hot spot identification method based on deep stack type hybrid self-coding network | |
| CN118736362A (en) | Interpretable image text description method based on multi-model fusion | |
| CN118261902A (en) | Lightweight multi-branch YOLO model for insulator defect detection and detection method thereof |
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 |