CN116452963A - Multi-dimensional feature-based seabed substrate type classification method - Google Patents
Multi-dimensional feature-based seabed substrate type classification method Download PDFInfo
- Publication number
- CN116452963A CN116452963A CN202310024192.9A CN202310024192A CN116452963A CN 116452963 A CN116452963 A CN 116452963A CN 202310024192 A CN202310024192 A CN 202310024192A CN 116452963 A CN116452963 A CN 116452963A
- Authority
- CN
- China
- Prior art keywords
- features
- substrate
- seabed
- image
- feature
- 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
- 239000000758 substrate Substances 0.000 title claims abstract description 63
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000013528 artificial neural network Methods 0.000 claims abstract description 17
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 17
- 238000012545 processing Methods 0.000 claims abstract description 5
- 230000009467 reduction Effects 0.000 claims abstract description 4
- 210000002569 neuron Anatomy 0.000 claims description 10
- 238000007637 random forest analysis Methods 0.000 claims description 4
- 238000012706 support-vector machine Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000003066 decision tree Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 abstract description 5
- 238000000605 extraction Methods 0.000 abstract description 3
- 238000012876 topography Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 5
- 238000009826 distribution Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000007636 ensemble learning method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000013049 sediment Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/05—Underwater scenes
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/54—Extraction of image or video features relating to texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention relates to a seabed substrate type classification method based on multidimensional features, which comprises the following steps: acquiring a seabed substrate image by adopting image sonar and extracting multidimensional features; adopting a Relieff algorithm to perform dimension reduction processing on the multi-dimensional substrate image characteristics; and classifying the types of the seabed substrate by adopting a Stacking model of the BP neural network. According to the invention, the submarine image with abundant detail submarine topography and texture information is obtained through the image sonar, and on the basis of multi-dimensional feature extraction of time frequency, frequency domain, time frequency domain and the like, the characteristic selection is carried out on the extracted substrate image features by utilizing a reliefF algorithm, so that the correlation of classification features can be effectively reduced; the classifier with the optimized parameters is combined with the BP neural network by using the Stacking model, so that the function of classifying the seabed substrate types with stable performance can be realized, the accuracy of substrate classification is improved, and the problems of complex sampling, low efficiency, damage to the seabed environment and the like of the existing seabed substrate are solved.
Description
Technical Field
The invention belongs to the technical field of ocean exploration, relates to seabed substrate data processing, and particularly relates to a seabed substrate type classification method based on multidimensional features.
Background
The existing submarine substrate detection and classification mainly comprises a traditional station sampling method, underwater photographing and shooting observation, and analysis and classification after submarine substrate data are acquired by utilizing an acoustic and electromagnetic wave indirect remote sensing mode, wherein the analysis and classification methods are different in operation means and principles.
The existing station sampling method mainly utilizes tools (water weight, grab bucket, sampler and the like) to directly sample the seabed substrate to determine the type and distribution of the substrate. Although the method is accurate and visual on the whole, the method has the defects of heavy equipment, low efficiency, high cost and the like, and is not suitable for large-area substrate research and investigation. In addition, the submarine environment can be damaged in the sampling process, the composition components in the substrate sample can be influenced, and adverse interference is brought to the identification of the substrate type.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a multi-dimensional characteristic-based seabed substrate type classification method, and solves the problems of complex sampling, low efficiency, damage to seabed environment and the like of the prior seabed substrate.
The invention solves the technical problems in the prior art by adopting the following technical scheme:
a method for classifying seabed sediment types based on multidimensional characteristics, comprising the following steps:
step 1, acquiring a seabed substrate image by adopting image sonar and extracting multidimensional features;
step 2, adopting a ReliefF algorithm to perform dimension reduction processing on the multi-dimensional substrate image characteristics;
and step 3, classifying the types of the seabed substrate by adopting a Stacking model of the BP neural network.
Further, the substrate image features extracted by the multi-dimensional features in the step 1 comprise statistical features, texture features and image frequency domain features.
Further, the statistical features include mean, standard deviation, kurtosis and skewness, and the calculation formulas are respectively as follows:
where Mean is the Mean, std is the standard deviation, kur is the kurtosis, ske is the skewness, and m and n represent the length and width of the image, respectively.
Further, the frequency domain features include: FFT mean, FFT variance, and wavelet packet decomposition band component energy ratio.
Further, the texture feature comprises: contrast, correlation, energy, homogeneity of GLCM in the 0 °, 45 °, 90 ° and 135 ° directions; short run-length factor of GLRLM in 0 ° direction; roughness, contrast and orientation of Tumura texture features; krawtchouk moment features; LBP features.
Further, the ReliefF algorithm in the step 2 updates the importance of a feature of all selected samples by calculating the relation value between the feature and the selected sample, and the specific implementation method comprises the following steps:
setting the importance WA of all feature A as 0;
randomly selecting one sample R from all sample data i At the same time, k samples closest to the sample data in the same class are selected to be denoted as H, H= { H j ,j＝1,2,...,k}；
Thirdly, k and samples R are respectively selected from the data samples of other different categories C i The nearest sample is denoted as M (C), M (C) = { M j (C) J=1, 2,..k }, where c+.class (R i )；
And fourthly, updating the feature importance within the set iteration times m, and finally judging and eliminating redundant features through a threshold value.
Further, the step of updating the feature importance is performed according to the following formula:
wherein P (class (R) i ) Is sample R) i The class samples occupy the whole sample proportion, and P (C) is the proportion of the class C samples to the whole sample proportion.
Further, the Stacking model in the step 3 includes two parts, namely a primary learning layer and a secondary learning layer, wherein the primary learning layer includes n different base classifiers, and the secondary learning layer is a meta classifier; the base classifier adopts a support vector machine, a random forest and a gradient lifting decision tree, the prediction probability fusion mode adopts a probability splicing mode, the element classifier adopts a BP neural network, and the BP neural network adjusts the weight and the threshold value of each neuron through a back propagation algorithm.
The invention has the advantages and positive effects that:
according to the invention, the submarine image with abundant submarine topography and texture information is obtained through the image sonar, so that the submarine substrate characteristics can be effectively reflected, and the correlation of the classification characteristics can be effectively reduced by adopting a multi-dimensional characteristic submarine substrate classification method and performing characteristic selection on the extracted substrate image characteristics by utilizing a Relieff algorithm on the basis of multi-dimensional characteristic extraction of time frequency, frequency domain, time frequency domain and the like; the classifier with the optimized parameters is combined with the BP neural network by using the Stacking model, so that the function of classifying the seabed substrate types with stable performance can be realized, the seabed substrate classification method is perfected, and the accuracy of substrate classification is improved.
Drawings
FIG. 1 is a schematic diagram of a Satcking model of the present invention;
FIG. 2 is a BP neural network topology of the present invention;
FIG. 3 is a neuron model of a BP neural network of the present invention;
FIG. 4 is a Stacking model structure employed in the present invention;
FIG. 5 is a graph comparing substrate distributions predicted by different classification methods.
Detailed Description
Embodiments of the present invention are described in further detail below with reference to the accompanying drawings.
The invention provides a seabed substrate type classification method based on multidimensional features, which comprises the following steps:
and step 1, acquiring a seabed substrate image by adopting image sonar, and extracting multidimensional features.
The feature of the substrate image is an important basis for training a classification model to realize substrate classification, but the information of the substrate image is difficult to comprehensively describe only by means of the intensity information of the substrate image, so that the multi-angle and multi-scale feature description of the substrate image is needed. In the invention, the extracted substrate image features mainly comprise statistical features, texture features and image frequency domain features.
The statistical features have the characteristics of simple extraction and clear physical meaning. The gray value description and the multi-order statistical moment of the image can indirectly reflect the change rule and the distribution rule of the gray value, and have certain classification performance. The Mean (Mean), standard deviation (Std), kurtosis (Kur) and skewness (Ske) of the substrate image are extracted as statistical classification features of the substrate image. Under the condition that the image data block size is m×n, the calculation formulas of the four image gray-scale statistical features are shown as formula (1) -formula (4), respectively.
In addition, in order to fully reflect the substrate type difference, the invention extracts statistical characteristics, frequency domain characteristics and texture characteristics, as shown in the following table.
TABLE 1 characterization of the method
And 2, performing dimension reduction processing on the multi-dimensional substrate image characteristics by adopting a ReliefF algorithm.
Certain correlation may exist between the substrate image features extracted in the step 1, which not only brings about huge classifier training time and substrate classification time consumption, but also affects classification accuracy. In order to eliminate redundancy and correlation between extracted image features, reduce feature dimensions, retain representative features with excellent classification performance, improve the efficiency of substrate classification, and select the final effective classification features by using a proper feature selection algorithm to perform feature selection on the extracted features.
The ReliefF algorithm is an improved version of the Relief algorithm, overcomes the limitation that the Relief algorithm can only perform feature selection on two classification problems, can perform feature selection on multiple classification problems, is high in operation speed and high in robustness, and can process incomplete and noisy data. The ReliefF algorithm updates the importance of a feature by calculating the relation value of the feature of all selected samples and the selected samples, and the higher the feature importance is, the better the classification performance of the feature is. The method for calculating the feature importance by the Relieff algorithm and realizing feature selection comprises the following steps:
(1) Setting the importance WA of all feature A as 0;
(2) Randomly selecting one sample R from all sample data i At the same time, k samples closest to the sample data in the same class are selected to be denoted as H, H= { H j ,j＝1,2,...,k}；
(3) In other different categories C (C. Noteq. Class (R) i ) K and samples R are selected from the data samples i The nearest sample is denoted as M (C), M (C) = { M j (C),j＝1,2,...,k}；
(4) And in the set iteration times m, updating the importance of the features through formulas (5) and (6), and finally judging and eliminating redundant features through a threshold value.
Wherein P (class (R) i ) Is sample R) i The class samples occupy the whole sample proportion, and P (C) is the proportion of the class C samples to the whole sample proportion.
And step 3, classifying the types of the seabed substrate by adopting a Stacking model of the BP neural network.
In the step, classification is realized by adopting a Stacking model ensemble learning method, and classification performance is further improved while overfitting is avoided by utilizing the advantages of each classifier model.
As shown in fig. 1, the Stacking model is composed of a primary learning layer and a secondary learning layer, where the primary learning layer mainly includes n different classifier models (base classifier), and the secondary learning layer is usually a simple classifier model (meta classifier).
In the invention, a Support Vector Machine (SVM), a Random Forest (RF) and a gradient lifting decision tree (GBDT) are selected as the base classifier, a probability splicing mode is adopted as a prediction probability fusion mode, and a BP neural network is adopted as the element classifier. The BP neural network is a topological network structure for training the neuron weight of the network by back propagation of errors, and mainly comprises an input layer, a hidden layer and an output layer, wherein the topological structure and the neuron model of the BP neural network are respectively shown in fig. 2 and 3.
Wherein { x i I=1, 2,..n } is the input to the neuron, { w i I=1, 2,..n } is the weight corresponding to each input, θ is the neuron threshold, f (·) is the neuron activation function, typically the sigmoid function or the tanh function, y is the neuron output, i.e.:
the whole BP neural network adjusts the weights and the thresholds of the neurons through a back propagation algorithm. Finally, the Stacking model of the invention combined with the BP neural network is shown in figure 4.
In order to verify the performance of the invention, the method is adopted to classify the seabed substrate, and the substrate distribution predicted by each classification method is shown in fig. 5 by comparing with other classification methods, so that the substrate classification function can be effectively realized by adopting the invention (shown as h in the figure).
It should be emphasized that the examples described herein are illustrative rather than limiting, and therefore the invention includes, but is not limited to, the examples described in the detailed description, all other embodiments that may be derived by a person skilled in the art from the technical solutions of the invention are equally within the scope of the invention.
Claims (8)
1. A kind of classification method of seabed substrate based on multidimensional characteristic, characterized by: the method comprises the following steps:
step 1, acquiring a seabed substrate image by adopting image sonar and extracting multidimensional features;
step 2, adopting a ReliefF algorithm to perform dimension reduction processing on the multi-dimensional substrate image characteristics;
and step 3, classifying the types of the seabed substrate by adopting a Stacking model of the BP neural network.
2. The multi-dimensional feature-based seabed substrate type classification method as claimed in claim 1, wherein: and the substrate image features extracted by the multi-dimensional features in the step 1 comprise statistical features, texture features and image frequency domain features.
3. A method of classifying a type of seabed substrate based on multi-dimensional features as claimed in claim 2, wherein: the statistical characteristics comprise mean value, standard deviation, kurtosis and skewness, and the calculation formulas are as follows:
where Mean is the Mean, std is the standard deviation, kur is the kurtosis, ske is the skewness, and m and n represent the length and width of the image, respectively.
4. A method of classifying a type of seabed substrate based on multi-dimensional features as claimed in claim 2, wherein: the frequency domain features include: FFT mean, FFT variance, and wavelet packet decomposition band component energy ratio.
5. A method of classifying a type of seabed substrate based on multi-dimensional features as claimed in claim 2, wherein: the texture features include: contrast, correlation, energy, homogeneity of GLCM in the 0 °, 45 °, 90 ° and 135 ° directions; short run-length factor of GLRLM in 0 ° direction; roughness, contrast and orientation of Tumura texture features; krawtchouk moment features; LBP features.
6. The multi-dimensional feature-based seabed substrate type classification method as claimed in claim 1, wherein: in the step 2, the ReliefF algorithm updates the importance of a certain feature of all selected samples by calculating the relation value of the feature and the selected samples to perform feature selection, and the specific implementation method comprises the following steps:
setting the importance WA of all feature A as 0;
randomly selecting one sample R from all sample data i At the same time, k samples closest to the sample data in the same class are selected to be denoted as H, H= { H j ,j＝1,2,...,k}；
Thirdly, k and samples R are respectively selected from the data samples of other different categories C i The nearest sample is denoted as M (C), M (C) = { M j (C) J=1, 2,..k }, where c+.class (R i )；
And fourthly, updating the feature importance within the set iteration times m, and finally judging and eliminating redundant features through a threshold value.
7. The multi-dimensional feature-based seabed substrate type classification method as claimed in claim 6, wherein: and step four, updating the feature importance according to the following formula:
wherein P (class (R) i ) Is sample R) i The class samples occupy the whole sample proportion, and P (C) is the proportion of the class C samples to the whole sample proportion.
8. The multi-dimensional feature-based seabed substrate type classification method as claimed in claim 1, wherein: the Stacking model in the step 3 comprises a primary learning layer and a secondary learning layer, wherein the primary learning layer comprises n different base classifiers, and the secondary learning layer is a meta classifier; the base classifier adopts a support vector machine, a random forest and a gradient lifting decision tree, the prediction probability fusion mode adopts a probability splicing mode, the element classifier adopts a BP neural network, and the BP neural network adjusts the weight and the threshold value of each neuron through a back propagation algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310024192.9A CN116452963A (en) | 2023-01-09 | 2023-01-09 | Multi-dimensional feature-based seabed substrate type classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310024192.9A CN116452963A (en) | 2023-01-09 | 2023-01-09 | Multi-dimensional feature-based seabed substrate type classification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116452963A true CN116452963A (en) | 2023-07-18 |
Family
ID=87132608
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310024192.9A Pending CN116452963A (en) | 2023-01-09 | 2023-01-09 | Multi-dimensional feature-based seabed substrate type classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116452963A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116863343A (en) * | 2023-09-04 | 2023-10-10 | 中国地质大学（武汉） | Deep learning model, seabed substrate interpretation method, device and medium |
-
2023
- 2023-01-09 CN CN202310024192.9A patent/CN116452963A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116863343A (en) * | 2023-09-04 | 2023-10-10 | 中国地质大学（武汉） | Deep learning model, seabed substrate interpretation method, device and medium |
| CN116863343B (en) * | 2023-09-04 | 2024-01-23 | 中国地质大学（武汉） | Deep learning model, seabed substrate interpretation method, device and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109086824B (en) | Seabed substrate sonar image classification method based on convolutional neural network | |
| CN109993280B (en) | Underwater sound source positioning method based on deep learning | |
| CN109597043B (en) | Radar signal identification method based on quantum particle swarm convolutional neural network | |
| WO2021128510A1 (en) | Bearing defect identification method based on sdae and improved gwo-svm | |
| CN109828251B (en) | Radar target identification method based on characteristic pyramid light-weight convolution neural network | |
| CN111161207B (en) | Integrated convolutional neural network fabric defect classification method | |
| CN111709315A (en) | Underwater acoustic target radiation noise identification method based on field adaptation | |
| CN114595732B (en) | Radar radiation source sorting method based on depth clustering | |
| CN116452963A (en) | Multi-dimensional feature-based seabed substrate type classification method | |
| CN103077408A (en) | Method for converting seabed sonar image into acoustic substrate classification based on wavelet neutral network | |
| CN115170943A (en) | Improved visual transform seabed substrate sonar image classification method based on transfer learning | |
| Ye et al. | Recognition algorithm of emitter signals based on PCA+ CNN | |
| CN113076920B (en) | Intelligent fault diagnosis method based on asymmetric domain confrontation self-adaptive model | |
| CN111882554B (en) | SK-YOLOv 3-based intelligent power line fault detection method | |
| CN108763926A (en) | A kind of industrial control system intrusion detection method with security immunization ability | |
| Ma | Improving SAR target recognition performance using multiple preprocessing techniques | |
| CN112508106A (en) | Underwater image classification method based on convolutional neural network | |
| CN113523904A (en) | Cutter wear detection method | |
| CN116680639A (en) | Deep-learning-based anomaly detection method for sensor data of deep-sea submersible | |
| CN113723482B (en) | Hyperspectral target detection method based on multi-example twin network | |
| Chen et al. | A review of underwater target recognition based on deep learning | |
| CN111783526B (en) | Cross-domain pedestrian re-identification method using posture invariance and graph structure alignment | |
| CN114444544A (en) | Signal classification and identification method based on convolutional neural network and knowledge migration | |
| Bin et al. | Automatic inspection of surface crack in solar cell images | |
| Tao et al. | Modulation classification based on multifractal features |
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 |