CN112926380A - Novel underwater laser target intelligent recognition system - Google Patents
Novel underwater laser target intelligent recognition system Download PDFInfo
- Publication number
- CN112926380A CN112926380A CN202110023538.4A CN202110023538A CN112926380A CN 112926380 A CN112926380 A CN 112926380A CN 202110023538 A CN202110023538 A CN 202110023538A CN 112926380 A CN112926380 A CN 112926380A
- Authority
- CN
- China
- Prior art keywords
- dimensional
- module
- image
- sample
- dimensional image
- 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.)
- Granted
Links
- 238000000605 extraction Methods 0.000 claims abstract description 21
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 14
- 230000004927 fusion Effects 0.000 claims abstract description 12
- 239000011159 matrix material Substances 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 8
- 238000004458 analytical method Methods 0.000 claims description 7
- 238000001914 filtration Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 3
- 125000004122 cyclic group Chemical group 0.000 claims description 2
- 238000003702 image correction Methods 0.000 claims description 2
- 230000003993 interaction Effects 0.000 claims description 2
- 238000003062 neural network model Methods 0.000 claims 1
- 238000001514 detection method Methods 0.000 description 10
- 238000000034 method Methods 0.000 description 10
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000000306 recurrent effect Effects 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 2
- 230000001902 propagating effect Effects 0.000 description 2
- 241000282414 Homo sapiens Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 239000003643 water by type Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a novel underwater laser target intelligent identification system which comprises an underwater laser radar, a three-dimensional sample library, a two-dimensional image database and an underwater laser target identification upper computer system, wherein the underwater laser target identification upper computer system comprises a sample distributed image generation module, a feature extraction module, a convolution attention module, a feature selection module, a feature fusion module, a decision module, an output and display module and the like. And provides an algorithm for carrying out target identification by using the underwater three-dimensional stereo sample. The invention has the advantages of realizing the extraction of a plurality of images from a three-dimensional sample and the effective characteristic extraction, identifying the moving target object in a turbid water area, having high identification accuracy and high intelligent degree, and simultaneously solving the problems of characteristic loss, high noise of the image in the turbid water area, identification of the number of small samples of the moving target and the like when a three-dimensional sample is converted into a two-dimensional image, which are commonly existed in the traditional underwater laser identification system.
Description
Technical Field
The invention relates to the fields of computer vision, intelligent recognition algorithm, laser radar and underwater target recognition, in particular to a novel underwater target intelligent recognition technology based on laser radar.
Background
The ocean area accounts for more than seven percent of the total area of the earth and has very much resources and space, but the ocean space which is explored by human beings currently is only about 5 percent of the total space of the earth, the ocean has important military strategic position and resource exploitation and utilization position, and due to the fact that the ocean environment is inexperienced and changeable, people are used for exploring the ocean environment, and casualties can be caused with high probability. Therefore, the underwater world promotion schedule is explored with modern intelligent technology. At present, the commonly used method for acquiring underwater target information comprises acoustic detection and optical detection, and due to the difference of frequency and applicable wave band, the two detection modes are also commonly used in different detection scenes. The acoustic frequency is low, so that the method is commonly used for judging whether a target exists in a long distance, and the optical frequency is high, so that the method is commonly used for identifying a short-distance target.
Due to the strong attenuation of the water body to the incident light, the incident light presents exponential attenuation when propagating in water, and the common light source is difficult to achieve ideal detection depth when propagating in water due to poor collimation of the common light source. The appearance of laser provides a new underwater detection means, the low attenuation of the laser at an ocean window can reach the detection depth which is difficult to reach by the conventional optical means by utilizing the high collimation of the laser, particularly, the application of the conventional laser radar technology can not only carry out target detection at larger water depth, but also overcome the problem that the acoustic resistance of water and air is not matched, and the direct detection from water to underwater is realized. However, due to the complex underwater condition, the disordered motion of various organisms in water and a plurality of imaging noise points, an intelligent underwater recognition system is urgently needed to solve the problem of target recognition in image processing when a target needs to be accurately recognized in an underwater image obtained by laser.
The sample that laser radar drawed is three-dimensional stereo sample, because the target is often in the moving state under water, laser radar hardly gets a large amount of samples at this in-process of sample, consequently must adopt the small sample to carry out the analysis, and, under muddy complicated waters environment, visibility is extremely low, to the judgement of same object, we often need through different angles to three-dimensional sample carry out the projection to carry out characteristic's contrast and analysis with the two-dimensional image that the projection obtained, thereby obtain the judgement of generality to the discernment thing.
Disclosure of Invention
Aiming at the problem of mismatching of characteristic values when a three-dimensional sample is sampled into a two-dimensional image at present, the method solves the problems that an underwater target is in a moving state and a large sample is difficult to obtain in a snapshot mode, overcomes the problem of insufficient imaging definition in a turbid water area,
the technical scheme adopted by the invention for solving the technical problems is as follows: the novel underwater laser target intelligent identification system is characterized by comprising an underwater laser radar, a three-dimensional sample library, a two-dimensional image database and an underwater laser target identification upper computer system, wherein the underwater laser target identification upper computer system comprises a sample distributed image generation module, a convolution attention module, a feature extraction module, a feature selection module, a feature fusion module, a decision module, an output and display module and the like. The method is characterized in that an underwater laser radar performs laser scanning on a target recognition object in a selected water area, three-dimensional sample data returned by the radar is stored in a three-dimensional sample bank, a sample distributed image generation module in an upper computer system for underwater laser target recognition performs automatic multi-angle shooting on the three-dimensional sample data in the three-dimensional sample bank to obtain two-dimensional image samples, data enhancement is performed on the two-dimensional image samples, then the processed two-dimensional image samples are stored in a two-dimensional image database, a feature extraction module performs feature extraction on the enhanced two-dimensional image and transfers the extracted features and the two-dimensional image to a convolution attention module, the convolution attention module automatically performs convolution coding on the features in a plurality of two-dimensional images from the same three-dimensional sample in the two-dimensional image database, and automatically establishes a mapping relation between a two-dimensional image space and a three-dimensional sample space, and the processed data is sent to a feature selection module, the feature selection module further screens the two-dimensional image features according to the attention coefficients generated in the convolutional codes, the feature fusion module further matches and corrects the screened features with the convolutional codes added with the attention coefficients after receiving the screened features, the features are delivered to a decision module for judgment of target recognition, and the decision module sends the final decision to an output and display module for a man-machine interaction and display link.
Furthermore, the sample distributed image generation module can automatically take two-dimensional image samples from multiple angles for three-dimensional sample data in the three-dimensional sample library, and perform image enhancement processing on the images, and the main processing algorithm is as follows:
2.1){xij,(i,j)∈I2i is a set or subset of natural numbers) to represent the gray value at each point of the digital image, where I represents the row position of the point in the lattice for which I represents the row position, j represents the column position of the point in the lattice for which j represents the column position, the gray value y at one point is replaced by the median of the point values in the neighborhood of that point: taking the gray value of the neighborhood of the point according to the size sequence x1≤x2≤x3≤...≤xnArrangement in which { x with a single numerical subscriptnN is a non-zero natural number } represents the label of the gray value of n points in the neighborhood, and then the intermediate value is taken according to the following mode:
2.2) processing the degraded image polluted by noise underwater by adopting a linear filtering mode, and completing the following steps: and the resulting gray value y of the two-dimensional median filter with the filter window AijCan be expressed as:
wherein (r, s) is the position information of the point in the filtering window, r is the abscissa of the position information, s is the ordinate of the position information,
indicating taking an intermediate value from each point gray value.
2.3) the calculation formula of the loss function L for carrying out the noise reduction operation on the image is as follows:
wherein i represents an image label ranging from 1 to N, N represents the number of training samples of the convolutional neural network of the image noise reduction module, and XiRepresenting the ith true noise-free picture, YiDenotes that the ith sheet is at XiAdding noise to the picture, R representing the original picture and YiThe residual pictures in between. The subscript F and superscript 2 in the formula indicate that the formula takes a 2 norm.
Further, the feature extraction module is used for analyzing the values of the RGB three-color channels and extracting features, and the mathematical basis for realizing the feature extraction module is as follows:
3.1) carrying out red channel compensation, wherein the compensation formula is as follows:
wherein, x is a certain pixel point in the image, alpha1In order to compensate for the coefficients of the coefficients,
is the average of the green color components,
is the mean of the red component, IR(x) Is the red component value, I, of the pointG(x) Is the green component value, I, of the pointd(x) To compensate for the red channel compensation value of the area where the pixel is located before,
the red channel compensation value of the area where the pixel is located after compensation.
3.2) to the image
Correction of color shift, maximum value per channel
And minimum value
The definition formula is:
3.3) HSV equalization and deblurring, gamma correction of the luminance, hue and saturation components to enhance hue and luminance contrast, and deblurring of the luminance.
3.4) carrying out feature extraction on the image.
Furthermore, the convolution attention module is used for carrying out convolution coding on the characteristics of a plurality of two-dimensional images from the same three-dimensional sample, generating an attention coefficient, and automatically establishing a mapping relation between a two-dimensional image space and a three-dimensional sample space so as to solve the problem of characteristic mismatching caused by shooting the two-dimensional images at various angles. The mathematical principle expression is as follows:
in the formula Ic(x) Is a three-dimensional sample function, x is a certain pixel point in the image, Jc(x) For a picture in a two-dimensional image function set, c ═ R, G, B },
as background light or backscattered light, tc(x) The attention coefficient corresponding to the two-dimensional image. By transforming the formula, a two-dimensional image function J is obtainedc(x) Comprises the following steps:
further, the feature selection module and the feature fusion module adopt a long-time memory recurrent neural network (LSTM-RNN) model to perform feature selection and feature fusion, and the state expression at the time t is as follows:
st＝f(Uxt+Wst-1)
ht＝g(Vst)
where t denotes the time, stIndicating the hidden layer state at time t, st-1Indicating the state of the hidden layer at the unit time immediately preceding time t. h istIndicating output layer h at time ttStatus. The hidden layer activation function is f, and U is expressed as the weight between the input layer and the hidden layerAnd W is a weight matrix between the hidden layer and the hidden layer from the time t-1 to the time t, and V is a weight matrix between the output layer and the hidden layer.
Further, the decision module is used for analyzing and fitting the extracted features and finally transmitting the analysis result to the output and display module, and the mathematical support of the decision module is as follows:
6.1) output layer weight equation
6.2) equation of weights of Loop layers
6.3) input layer weight equation
Wherein, why(i, j) represents a weight matrix from the hidden layer to the output layer, w'hyAnd (i, j) represents a weight matrix from the hidden layer to the output layer after the weight is updated. w is ahh(i, j) represents a weight matrix between hidden layers, w'hhAnd (i, j) represents a weight matrix between hidden layers after the weight is updated. w is axh(i, j) represents a weight matrix, w ', between the input layer and the hidden layer'xhAnd (i, j) represents a weight matrix between the input layer and the hidden layer after the weight is updated. Eta is expressed as a learning rate,
is the output layer error.
The technical conception of the invention is as follows: aiming at the problems that the moving frequency of a current underwater target is high, and the number of three-dimensional space samples obtained by a laser radar is small, a sample distributed image generation module is designed to perform projection analysis on three-dimensional space small samples from different angles. Aiming at the problem that a traditional neural network structure cannot link the same characteristics of a plurality of pictures for learning in the process of projecting a three-dimensional space to a two-dimensional screen, a convolution attention module is introduced to carry out convolution coding on the characteristics from the three-dimensional space to the two-dimensional space, the mapping relation between the two-dimensional image space and the three-dimensional sample space is automatically established, and a long-time memory recurrent neural network (LSTM-RNN) model neural network is set up for learning. Aiming at the phenomena of low visibility under turbid water and serious noise of the obtained image, a feature extraction mode of red channel enhancement is adopted, and the feature extraction mode is improved.
The invention has the following beneficial effects: 1. the method combines the traditional CNN and the characteristic fusion mode, provides a new technical scheme applied to the underwater target identification of three-dimensional small sample number, and has the advantages of lower detection cost, higher efficiency and better effect compared with the existing underwater target identification technical scheme. 2. Based on an attention mechanism, convolution attention is introduced into the established long-time memory cyclic neural network, the relation between the characteristic channel and the space is enhanced, and the relation between the two-dimensional image and the three-dimensional space is established. 3. A large number of experiments prove that the method is high in efficiency and stability, and compared with a common underwater target recognition system, the method is high in recognition accuracy and high in feature analysis efficiency.
Drawings
Fig. 1 is a structural model diagram of the system proposed by the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings.
Referring to fig. 1, the underwater laser radar 1, the three-dimensional sample library 2, the two-dimensional image database 3 and the underwater laser target recognition upper computer system 4 are formed, and the underwater laser target recognition upper computer system 4 is formed by sequentially connecting a sample distributed image generation module 5, a convolution attention module 6, a feature extraction module 7, a feature selection module 8, a feature fusion module 9, a decision module 10 and an output and display module 11.
The underwater laser radar 1 obtains a three-dimensional sample of a target object to be detected by emitting laser and receiving the laser reflected by the target, and stores the three-dimensional sample into a three-dimensional sample library 2.
A sample distributed image generation module 5 in the underwater laser target recognition upper computer system automatically shoots three-dimensional sample data in a three-dimensional sample library 2 from multiple angles to obtain two-dimensional image samples, performs data enhancement on the two-dimensional image samples, and then stores the processed two-dimensional image samples into a two-dimensional image database 3. In this step, the image enhancement processing on the image sample is processed according to the following formula:
{xij,(i,j)∈I2i is a set or subset of natural numbers) to represent the gray value at each point of the digital image, where I represents the row position of the point in the lattice for which I represents the row position, j represents the column position of the point in the lattice for which j represents the column position, the gray value y at one point is replaced by the median of the point values in the neighborhood of that point: taking the gray value of the neighborhood of the point according to the size sequence x1≤x2≤x3≤...≤xnArrangement in which { x with a single numerical subscriptnN is a non-zero natural number } represents the label of the gray value of n points in the neighborhood, and then the intermediate value is taken according to the following mode:
processing the degraded image polluted by noise underwater by adopting a linear filtering mode, and completing the following steps: and the resulting gray value y of the two-dimensional median filter with the filter window AijCan be expressed as:
wherein (r, s) is the position information of the point in the filtering window, r is the abscissa of the position information, s is the ordinate of the position information,
indicating taking an intermediate value from each point gray value. To proceed with imageThe loss function L of the noise reduction operation is calculated as follows:
wherein i represents an image label ranging from 1 to N, N represents the number of training samples of the convolutional neural network of the image noise reduction module, and XiRepresenting the ith true noise-free picture, YiDenotes that the ith sheet is at XiAdding noise to the picture, R representing the original picture and YiThe residual pictures in between. The subscript F and superscript 2 in the formula indicate that the formula takes a 2 norm.
The feature extraction module 6 is used for analyzing the RGB three-color channel values, extracting features, and compensating the red channel, wherein the compensation formula is as follows:
wherein, x is a certain pixel point in the image, alpha1In order to compensate for the coefficients of the coefficients,
is the average of the green color components,
is the mean of the red component, IR(x) Is the red component value, I, of the pointG(x) Is the green component value, I, of the pointd(x) To compensate for the red channel compensation value of the area where the pixel is located before,
the red channel compensation value of the area where the pixel is located after compensation. For images
Correction of color shift, maximum value per channel
And minimum value
The definition formula is:
HSV equalization and deblurring are adopted, gamma correction is carried out on the brightness, hue and saturation components to enhance hue and brightness contrast, deblurring is carried out on the brightness, and feature extraction is carried out on the image.
The convolution attention module 7 is used for performing convolution coding on the characteristics of a plurality of two-dimensional images from the same three-dimensional sample, generating an attention coefficient, and automatically establishing a mapping relation between a two-dimensional image space and a three-dimensional sample space so as to solve the problem of characteristic mismatching caused by shooting the two-dimensional images at various angles. The mathematical principle expression is as follows:
in the formula Ic(x) Is a three-dimensional sample function, x is a certain pixel point in the image, Jc(x) For a picture in a two-dimensional image function set, c ═ R, G, B },
as background light or backscattered light, tc(x) The attention coefficient corresponding to the two-dimensional image. By transforming the formula, a two-dimensional image function J is obtainedc(x) Comprises the following steps:
the feature selection module 8 and the feature fusion module 9 adopt a long-time memory recurrent neural network (LSTM-RNN) model to perform feature selection and feature fusion, and the state expression at the time t is as follows:
st＝f(Uxt+Wst-1)
ht＝g(Vst)
where t denotes the time, stIndicating the hidden layer state at time t, st-1Indicating the state of the hidden layer at the unit time immediately preceding time t. h istIndicating output layer h at time ttStatus. The hidden layer activation function is f, U is expressed as a weight matrix between the input layer and the hidden layer, W is expressed as a weight matrix between the hidden layer and the hidden layer from the time t-1 to the time t, and V is expressed as a weight matrix between the output layer and the hidden layer.
The decision module 10 is used for analyzing and fitting the extracted features and finally transmitting the analysis result to the output and display module 11, and the mathematical support of the decision module is as follows:
output layer weight equation
Circulating layerWeight equation
Input layer weight equation
Wherein, why(i, j) represents a weight matrix from the hidden layer to the output layer, w'hyAnd (i, j) represents a weight matrix from the hidden layer to the output layer after the weight is updated. w is ahh(i, j) represents a weight matrix between hidden layers, w'hhAnd (i, j) represents a weight matrix between hidden layers after the weight is updated. w is axh(i, j) represents a weight matrix, w ', between the input layer and the hidden layer'xhAnd (i, j) represents a weight matrix between the input layer and the hidden layer after the weight is updated. Eta is expressed as a learning rate,
is the output layer error.
The output and display module 11 displays the final result of the classification recognition of the image features.
The above examples are intended to illustrate the invention, but not to limit the invention, and any modifications and variations of the invention within the spirit of the invention and the scope of the claims are within the scope of the invention.
Claims (6)
1. The novel underwater laser target intelligent identification system is characterized by comprising an underwater laser radar, a three-dimensional sample library, a two-dimensional image database and an underwater laser target identification upper computer system, wherein the underwater laser target identification upper computer system comprises a sample distributed image generation module, a convolution attention module, a feature extraction module, a feature selection module, a feature fusion module, a decision module, an output and display module and the like. The underwater laser radar performs laser scanning on a target recognition object in a selected water area, stores three-dimensional sample data returned by the radar into a three-dimensional sample library, a sample distributed image generation module performs multi-angle shooting on the three-dimensional sample data in the three-dimensional sample library to obtain two-dimensional image samples, performs data enhancement on the two-dimensional image samples, then stores the processed two-dimensional image samples into a two-dimensional image database, a feature extraction module performs feature extraction on the enhanced two-dimensional image and transfers the extracted features and the two-dimensional image to a convolution attention module, the convolution attention module performs convolution coding on the features in a plurality of two-dimensional images from the same three-dimensional sample in the two-dimensional image database, automatically establishes a mapping relation between a two-dimensional image space and a three-dimensional sample space, and sends the processed data to a feature selection module, the feature selection module further screens the two-dimensional image features according to the attention coefficients generated in the convolutional coding; after receiving the screened features, the feature fusion module further matches and corrects the screened features with the convolution codes added with attention coefficients, and then the selected features are sent to a decision module for judgment of target identification; the decision module sends the final decision to the display module for human-computer interaction and display.
2. The novel underwater laser target intelligent identification system according to claim 1, wherein the sample distributed image generation module is capable of performing multi-angle shooting of two-dimensional image samples on three-dimensional sample data in a three-dimensional sample library, and performing image enhancement processing on the images, and specifically comprises:
(2.1){xij,(i,j)∈I2i is a set or subset of natural numbers) to represent the gray value at each point of the digital image, where I represents the row position of the point in the lattice for which I represents the row position, j represents the column position of the point in the lattice for which j represents the column position, the gray value y at one point is replaced by the median of the point values in the neighborhood of that point: taking the gray value of the neighborhood of the point according to the size sequence x1≤x2≤x3≤...≤xnArrangement in which { x with a single numerical subscriptnN is a non-zero natural number } represents the label of the gray value of n points in the neighborhood, and then the intermediate value is taken according to the following mode:
(2.2) processing the degraded image polluted by noise underwater by adopting a linear filtering mode, and completing the following steps: and the resulting gray value y of the two-dimensional median filter with the filter window AijCan be expressed as:
wherein (r, s) is the position information of the point in the filtering window, r is the abscissa of the position information, s is the ordinate of the position information,
indicating taking an intermediate value from each point gray value.
(2.3) the loss function L for performing the noise reduction operation on the image is calculated as follows:
wherein i represents an image label ranging from 1 to N, N represents the number of training samples of the convolutional neural network of the image noise reduction module, and XiRepresenting the ith true noise-free picture, YiDenotes that the ith sheet is at XiAdding noise to the picture, R representing the original picture and YiThe residual pictures in between. The subscript F and superscript 2 in the formula indicate that the formula takes a 2 norm.
3. The system for intelligently identifying the underwater laser target according to claim 1, wherein the feature extraction module is specifically configured to perform feature extraction on the enhanced two-dimensional image:
(3.1) carrying out red channel compensation, wherein the compensation formula is as follows:
wherein, x is a certain pixel point in the image, alpha1In order to compensate for the coefficients of the coefficients,
is the average of the green color components,
is the mean of the red component, IR(x) Is the red component value, I, of the pointG(x) Is the green component value, I, of the pointd(x) To compensate for the red channel compensation value of the area where the pixel is located before,
the red channel compensation value of the area where the pixel is located after compensation.
(3.2) for the image
Correction of color shift, maximum value per channel
And minimum value
The definition formula is:
(3.3) HSV equalization and deblurring, gamma correction of the luminance, hue and saturation components to enhance hue and luminance contrast, and deblurring of the luminance.
And (3.4) carrying out feature extraction on the image.
4. The system of claim 1, wherein the convolution attention module is configured to perform convolution encoding on a plurality of features of two-dimensional images from the same three-dimensional sample, generate an attention coefficient, and automatically establish a mapping relationship between a two-dimensional image space and a three-dimensional sample space, so as to solve a problem of feature mismatch caused by shooting two-dimensional images at various angles:
in the formula Ic(x) Is a three-dimensional sample function, x is a certain pixel point in the image, Jc(x) For a picture in a two-dimensional image function set, c ═ R, G, B },
as background light or backscattered light, tc(x) The attention coefficient corresponding to the two-dimensional image. By transforming the formula, a two-dimensional image function J is obtainedc(x) Comprises the following steps:
5. the system for intelligently identifying the underwater laser target according to claim 1, wherein the feature selection module and the feature fusion module adopt a long-time memory cyclic neural network model for feature extraction and selection, and a state expression at the time t is as follows:
st＝f(Uxt+Wst-1)
ht＝g(Vst)
where t denotes the time, stIndicating the hidden layer state at time t, st-1Indicating the state of the hidden layer at the unit time immediately preceding time t. h istIndicating output layer h at time ttStatus. The hidden layer activation function is f, U is expressed as a weight matrix between the input layer and the hidden layer, W is expressed as a weight matrix between the hidden layer and the hidden layer from the time t-1 to the time t, and V is expressed as a weight matrix between the output layer and the hidden layer.
6. The system for intelligently recognizing the underwater laser target according to claim 1, wherein the decision module is configured to analyze and fit the extracted features, and finally transmit the analysis result to the output and display module, and specifically:
(6.1) output layer weight equation
(6.2) equation of weights of Loop layers
(6.3) input layer weight equation
Wherein, why(i, j) represents a weight matrix from the hidden layer to the output layer, w'hyAnd (i, j) represents a weight matrix from the hidden layer to the output layer after the weight is updated. w is ahh(i, j) represents a weight matrix between hidden layers, w'hhAnd (i, j) represents a weight matrix between hidden layers after the weight is updated. w is axh(i, j) represents a weight matrix, w ', between the input layer and the hidden layer'xhAnd (i, j) represents a weight matrix between the input layer and the hidden layer after the weight is updated. Eta is expressed as a learning rate,
is the output layer error.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110023538.4A CN112926380B (en) | 2021-01-08 | 2021-01-08 | Novel underwater laser target intelligent recognition system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110023538.4A CN112926380B (en) | 2021-01-08 | 2021-01-08 | Novel underwater laser target intelligent recognition system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112926380A true CN112926380A (en) | 2021-06-08 |
| CN112926380B CN112926380B (en) | 2022-06-24 |
Family
ID=76162787
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110023538.4A Expired - Fee Related CN112926380B (en) | 2021-01-08 | 2021-01-08 | Novel underwater laser target intelligent recognition system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112926380B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115841614A (en) * | 2023-02-20 | 2023-03-24 | 中国石油大学(华东) | Image processing method and device, underwater imaging equipment and medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150262075A1 (en) * | 2012-06-22 | 2015-09-17 | Alex Nugent | Methods and systems for feature extraction of surface manifolds |
| CN105076077A (en) * | 2015-07-25 | 2015-11-25 | 孟宪胜 | Method for recognizing fish body based on laser imaging detection |
| CN108447074A (en) * | 2018-02-02 | 2018-08-24 | 中国科学院西安光学精密机械研究所 | Underwater target identification method based on bidirectional self-adaptive semantic fusion |
| CN109766811A (en) * | 2018-12-31 | 2019-05-17 | 复旦大学 | The end-to-end detection and recognition methods of sea ship in a kind of satellite-borne SAR image |
| CN110084234A (en) * | 2019-03-27 | 2019-08-02 | 东南大学 | A kind of sonar image target identification method of Case-based Reasoning segmentation |
| CN110334765A (en) * | 2019-07-05 | 2019-10-15 | 西安电子科技大学 | Remote Image Classification based on the multiple dimensioned deep learning of attention mechanism |
| CN110807365A (en) * | 2019-09-29 | 2020-02-18 | 浙江大学 | Underwater target identification method based on fusion of GRU and one-dimensional CNN neural network |
| CN111462131A (en) * | 2019-01-22 | 2020-07-28 | 斯特拉德视觉公司 | Method and equipment for attention-driven image segmentation |
| CN112163461A (en) * | 2020-09-07 | 2021-01-01 | 中国海洋大学 | Underwater target identification method based on multi-mode fusion |
-
2021
- 2021-01-08 CN CN202110023538.4A patent/CN112926380B/en not_active Expired - Fee Related
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150262075A1 (en) * | 2012-06-22 | 2015-09-17 | Alex Nugent | Methods and systems for feature extraction of surface manifolds |
| CN105076077A (en) * | 2015-07-25 | 2015-11-25 | 孟宪胜 | Method for recognizing fish body based on laser imaging detection |
| CN108447074A (en) * | 2018-02-02 | 2018-08-24 | 中国科学院西安光学精密机械研究所 | Underwater target identification method based on bidirectional self-adaptive semantic fusion |
| CN109766811A (en) * | 2018-12-31 | 2019-05-17 | 复旦大学 | The end-to-end detection and recognition methods of sea ship in a kind of satellite-borne SAR image |
| CN111462131A (en) * | 2019-01-22 | 2020-07-28 | 斯特拉德视觉公司 | Method and equipment for attention-driven image segmentation |
| CN110084234A (en) * | 2019-03-27 | 2019-08-02 | 东南大学 | A kind of sonar image target identification method of Case-based Reasoning segmentation |
| CN110334765A (en) * | 2019-07-05 | 2019-10-15 | 西安电子科技大学 | Remote Image Classification based on the multiple dimensioned deep learning of attention mechanism |
| CN110807365A (en) * | 2019-09-29 | 2020-02-18 | 浙江大学 | Underwater target identification method based on fusion of GRU and one-dimensional CNN neural network |
| CN112163461A (en) * | 2020-09-07 | 2021-01-01 | 中国海洋大学 | Underwater target identification method based on multi-mode fusion |
Non-Patent Citations (3)
| Title |
|---|
| WANG Z ET AL: "Attention-Aware Age-Agnostic Visual Place Recognition", 《2019 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION WORKSHOP（ICCVW）》 * |
| 王伟等: "基于注意机制和循环卷积神经网络的细粒度图像分类算法", 《西南师范大学学报(自然科学版)》 * |
| 王文海: "控制装备及***未来发展趋势", 《自动化博览》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115841614A (en) * | 2023-02-20 | 2023-03-24 | 中国石油大学(华东) | Image processing method and device, underwater imaging equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112926380B (en) | 2022-06-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112288658B (en) | Underwater image enhancement method based on multi-residual joint learning | |
| CN110363716B (en) | High-quality reconstruction method for generating confrontation network composite degraded image based on conditions | |
| CN112184577B (en) | Single image defogging method based on multiscale self-attention generation countermeasure network | |
| CN110349093B (en) | Single image defogging model construction and defogging method based on multi-stage hourglass structure | |
| CN111292264A (en) | Image high dynamic range reconstruction method based on deep learning | |
| CN110288550B (en) | Single-image defogging method for generating countermeasure network based on priori knowledge guiding condition | |
| Zong et al. | Local-CycleGAN: a general end-to-end network for visual enhancement in complex deep-water environment | |
| CN113222824B (en) | Infrared image super-resolution and small target detection method | |
| CN112562255A (en) | Intelligent image detection method for cable channel smoke and fire condition in low-light-level environment | |
| CN112926380B (en) | Novel underwater laser target intelligent recognition system | |
| CN112991218A (en) | Image processing method, device, equipment and storage medium | |
| CN113486894A (en) | Semantic segmentation method for satellite image feature component | |
| CN111539896A (en) | Domain-adaptive-based image defogging method and system | |
| Li et al. | Low-light underwater image enhancement for deep-sea tripod | |
| CN113298744B (en) | End-to-end infrared and visible light image fusion method | |
| CN111861949A (en) | Multi-exposure image fusion method and system based on generation countermeasure network | |
| CN113487530B (en) | Infrared and visible light fusion imaging method based on deep learning | |
| CN117115612A (en) | Multi-sensor image enhancement optimization-based multi-channel image fusion method and system | |
| Wang et al. | Underwater image quality optimization: Researches, challenges, and future trends | |
| CN113191971A (en) | Unmanned aerial vehicle image defogging method based on YUV color space | |
| CN114445288A (en) | Lightweight underwater image enhancement method based on deep learning | |
| CN113763356A (en) | Target detection method based on visible light and infrared image fusion | |
| CN111476353B (en) | Super-resolution method of GAN image introducing significance | |
| Martinho et al. | An Approach for Fish Detection in Underwater Images | |
| CN114373118B (en) | Underwater target detection method based on improved YOLOV4 |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220624 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |