CN118379611B - Ocean pollution early warning method based on image recognition - Google Patents
Ocean pollution early warning method based on image recognition Download PDFInfo
- Publication number
- CN118379611B CN118379611B CN202410790912.7A CN202410790912A CN118379611B CN 118379611 B CN118379611 B CN 118379611B CN 202410790912 A CN202410790912 A CN 202410790912A CN 118379611 B CN118379611 B CN 118379611B
- Authority
- CN
- China
- Prior art keywords
- area
- value
- lstm
- flocculent
- extraction unit
- 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
- 238000000034 method Methods 0.000 title claims abstract description 25
- 241001474374 Blennius Species 0.000 claims abstract description 76
- 230000008859 change Effects 0.000 claims abstract description 72
- 239000013535 sea water Substances 0.000 claims abstract description 52
- 238000013528 artificial neural network Methods 0.000 claims abstract description 16
- 230000004927 fusion Effects 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims abstract description 11
- 240000007058 Halophila ovalis Species 0.000 claims abstract 2
- 238000000605 extraction Methods 0.000 claims description 74
- 238000011176 pooling Methods 0.000 claims description 16
- 238000009499 grossing Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 abstract description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 abstract description 3
- 241000544058 Halophila Species 0.000 description 12
- 230000009286 beneficial effect Effects 0.000 description 8
- 230000006870 function Effects 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 241000251468 Actinopterygii Species 0.000 description 2
- 239000000356 contaminant Substances 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 241000579895 Chlorostilbon Species 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 229910052876 emerald Inorganic materials 0.000 description 1
- 239000010976 emerald Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000010842 industrial wastewater Substances 0.000 description 1
- 244000005700 microbiome Species 0.000 description 1
- 235000015097 nutrients Nutrition 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 230000029553 photosynthesis Effects 0.000 description 1
- 238000010672 photosynthesis Methods 0.000 description 1
- 230000000243 photosynthetic effect Effects 0.000 description 1
- 244000038651 primary producers Species 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 150000003839 salts Chemical class 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000035899 viability Effects 0.000 description 1
Classifications
-
- 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
Landscapes
- Image Analysis (AREA)
- Cultivation Of Seaweed (AREA)
Abstract
The invention discloses a marine pollution early warning method based on image recognition, which belongs to the technical field of image processing, and comprises the steps of acquiring a marine water bottom image, dividing the marine water bottom image according to color smooth values of pixel points to obtain a seaweed area, a sea water area and a flocculent area, calculating a marine pollution value according to the color values of the seaweed area, the color values of the sea water area and the concentration values of the flocculent area, so as to realize analysis of the current marine pollution condition, processing a sample by adopting a parallel LSTM enhanced characteristic fusion neural network, realizing comprehensive history of marine pollution values at all times, sea grass area change rate and flocculent area change rate, extracting characteristics of all time changes by adopting the parallel LSTM enhanced characteristic fusion neural network, improving the precision of predicting the marine future pollution value, and early warning marine pollution.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to an ocean pollution early warning method based on image recognition.
Background
Marine pollution refers to the phenomenon that various harmful substances enter the marine environment through different ways, so that the balance of a marine ecological system is broken and the biodiversity is damaged. With the accelerated development of industrialization and urbanization, the problem of marine pollution is increasingly serious, and the ecological environment and human health are greatly threatened.
Seaweed and floe are two typical pollution indicators during marine pollution. Seaweed is an important primary producer in the marine ecosystem, and they supply energy and oxygen to marine organisms through photosynthesis. However, when the sea is contaminated, the growth and reproduction of seaweed is affected, resulting in a reduced or even vanishing seaweed population.
Floc refers to a floc structure formed by aggregation of suspended particles in seawater, and is generally composed of microorganisms, organic matters, inorganic salts and the like. Flocs can be severely affected during marine pollution.
The existing ocean pollution early warning method based on image recognition mainly utilizes a remote sensing technology to acquire image information of the ocean surface, then processes and analyzes the image through a computer vision and deep learning algorithm to analyze the pollution condition of the ocean surface, but cannot analyze the pollution condition in the ocean and predict the future pollution condition of the ocean.
Disclosure of Invention
Aiming at the defects in the prior art, the marine pollution early warning method based on image recognition solves the problems that the prior art cannot analyze pollution conditions in the ocean and cannot predict future pollution conditions of the ocean.
In order to achieve the aim of the invention, the invention adopts the following technical scheme: an ocean pollution early warning method based on image recognition comprises the following steps:
S1, acquiring a submarine image in the ocean, and dividing the submarine image based on color smooth values of pixel points to obtain a seaweed area, a sea water area and a flocculent area;
s2, calculating a marine pollution value according to the color value of the seaweed area, the color value of the seawater area and the concentration value of the flocculent area;
s3, carrying out normalization processing on the ocean pollution value, the seaweed area change rate and the flocculent area change rate at each moment to obtain a sample;
s4, processing the sample by adopting a parallel LSTM enhanced feature fusion neural network to obtain a future ocean pollution value;
S5, early warning is carried out when the future pollution value of the ocean is larger than the pollution threshold value.
The beneficial effects of the invention are as follows: according to the method, a marine pollution value is calculated according to the color value of the sea grass area, the color value of the sea water area and the concentration value of the flocculent area, so that analysis of the current marine pollution condition is realized, a parallel LSTM enhanced characteristic fusion neural network is adopted to process samples, the marine future pollution value is predicted by comprehensively combining the marine pollution value, the sea grass area change rate and the flocculent area change rate at each moment, the marine pollution change condition is embodied through the marine pollution value at each moment, the sea grass change condition is embodied through the sea grass area change rate, the flocculent area change rate is embodied by the flocculent change condition, the characteristic of each moment change is extracted by adopting the parallel LSTM enhanced characteristic fusion neural network, the precision of predicting the marine future pollution value is improved, and the marine pollution is early-warned.
Further, the step S1 includes the following sub-steps:
S11, acquiring a submarine image in the ocean;
S12, denoising the underwater image to obtain an underwater denoising image;
S13, identifying a sea water area from the underwater denoising image according to the color smooth value of the pixel point;
s14, removing the sea water region from the underwater denoising image to obtain a plurality of independent closed subgraphs;
s15, identifying each independent closed subgraph to obtain a seaweed area and a flocculent area.
The beneficial effects of the above further scheme are: according to the method, after the underwater image in the ocean is acquired, denoising is carried out on the underwater image, the influence of noise points is removed, according to the color smoothing condition of each pixel point, a sea water area is firstly identified from the underwater denoising image, then the sea water area is removed from the underwater denoising image, and the rest area is segmented to obtain a sea weed area and a flocculent area.
Further, the step S13 includes the following sub-steps:
S131, calculating a color smooth value of each pixel point in the underwater denoising image;
s132, screening out pixel points with color smoothing values lower than a smoothing threshold value as candidate pixel points;
S133, clustering the candidate pixel points according to the color distance of each candidate pixel point to obtain a plurality of classifications;
S134, taking the classification with the largest number of candidate pixels as a sea water area.
Further, the formula for calculating the color smoothed value in S131 is:
，
Wherein S is a color smoothing value, R o is an R channel value of a pixel, G o is a G channel value of the pixel, B o is a B channel value of the pixel, R o,i is an ith R channel value of a neighborhood range of the pixel, G o,i is an ith G channel value of the neighborhood range of the pixel, B o,i is an ith B channel value of the neighborhood range of the pixel, I is an absolute value, N is the number of R channel values, G channel values or B channel values of the neighborhood range, and i is a positive integer.
The beneficial effects of the above further scheme are: according to the method, the color smooth value of each pixel point is calculated according to the conditions of a plurality of R channel values, G channel values and B channel values in the neighborhood range of each pixel point, the smaller the color smooth value is, the more uniform the color is, the larger the color change in a seaweed area and a flocculent area is, meanwhile, after candidate pixel points are selected, clustering processing is carried out on the candidate pixel points according to the color distance of each candidate pixel point, the largest classification is selected as a sea water area, and the interference of smooth areas in part of the seaweed area and the flocculent area on the sea water area is avoided.
Further, the step S15 includes the following sub-steps:
S151, taking the independent closed subgraph with the largest area as a seaweed area;
S152, selecting an independent closed sub-graph with the area smaller than an area threshold from the rest multiple independent closed sub-graphs as a flocculent area.
The beneficial effects of the above further scheme are: in S14, the sea water area is removed from the underwater denoising image, so that the rest area to be segmented is formed by a plurality of independent closed subgraphs, in the ocean, seaweed and the seabed are connected together, the area ratio of the sea grass in the image is large, the independent closed subgraphs with the largest area are used as the seaweed area, and the independent closed subgraphs with the area smaller than the area threshold are selected from the rest independent closed subgraphs, so that floccules distributed in the sea water are screened out, the independent closed subgraphs with the area larger than the area threshold are discarded, and the interference of fish is avoided.
Further, the calculation formula of the concentration value of the flocculent region in the S2 is as follows:
，
Wherein μ is a concentration value of a flocculent region, E F is an area of the flocculent region, and E S is an area of a seawater region;
the formula for calculating the ocean pollution value in the S2 is as follows:
，
wherein, gamma is the ocean pollution value, C G is the color value of the seaweed area, C S is the color value of the sea water area and is the color standard value of the sea grass area,Is the standard value of the color of the sea water area,As a standard value of the concentration of the flocculent region, alpha 1 is a first weighting coefficient, alpha 2 is a second weighting coefficient, and alpha 3 is a third weighting coefficient.
The beneficial effects of the above further scheme are: the invention synthesizes the difference between the color value of the sea weed area and the color standard value of the sea weed area, the difference between the color value of the sea water area and the color standard value of the sea water area, the difference between the concentration value of the flocculent area and the concentration standard value of the flocculent area, and obtains the ocean pollution value through weighting treatment, and whether the ocean is abnormal or not is reflected from three aspects.
Further, the calculation formula of the change rate of the seaweed area in the step S3 is as follows:
，
Wherein V G,t is the change rate of the seaweed area at the T moment, E G,t is the area of the seaweed area at the T moment, E G,t-1 is the area of the seaweed area at the T-1 moment, and T is the interval time length;
the calculation formula of the flocculent region change rate in the S3 is as follows:
，
Wherein V F,t is the change rate of the flocculent region at the t time, E F,t is the flocculent region area at the t time, E F,t-1 is the flocculent region area at the t-1 time, and t is the number of the time.
The beneficial effects of the above further scheme are: the invention calculates the change rate of the seaweed area and the change rate of the flocculent area, reflects whether the seaweed area has rapid change in a short time through the change rate of the seaweed area, reflects whether the flocculent area has rapid change in a short time through the change rate of the flocculent area, and can show that the environment in the ocean changes rapidly when the seaweed area and the flocculent area change rapidly.
Further, the parallel LSTM enhancement feature fusion neural network in S4 includes: the device comprises a first LSTM extraction unit, a second LSTM extraction unit, a third LSTM extraction unit, concat layers and a full-connection layer;
the input end of the first LSTM extraction unit is used for inputting ocean pollution values at all moments;
the input end of the second LSTM extraction unit is used for inputting the change rate of the seaweed area at each moment;
The input end of the third LSTM extraction unit is used for inputting the flocculent region change rate at each moment;
The input end of the Concat layers is respectively connected with the output end of the first LSTM extraction unit, the output end of the second LSTM extraction unit and the output end of the third LSTM extraction unit; the input end of the full-connection layer is connected with the output end of the Concat layers, and the output end of the full-connection layer is used as the output end of the parallel LSTM enhancement feature fusion neural network.
The beneficial effects of the above further scheme are: the invention inputs ocean pollution values at all times at a first LSTM extraction unit, captures the association condition of the ocean pollution values at all times, inputs the seaweed area change rate at all times at a second LSTM extraction unit, captures the association condition of the seaweed area change rate at all times, inputs the flocculent area change rate at all times at a third LSTM extraction unit, captures the association condition of the flocculent area change rate at all times, splices the output characteristics of three LSTM extraction units through Concat layers, and outputs the ocean future pollution values by adopting a full connection layer.
Further, the first LSTM extraction unit, the second LSTM extraction unit, and the third LSTM extraction unit each include: LSTM layer, salient feature enhancement layer and average pooling layer;
The input end of the LSTM layer is used as the input end of the first LSTM extraction unit, the second LSTM extraction unit or the third LSTM extraction unit, and the output end of the LSTM layer is connected with the input end of the salient feature enhancement layer; the input end of the average pooling layer is connected with the output end of the salient feature enhancement layer, and the output end of the average pooling layer is used as the output end of the first LSTM extraction unit, the second LSTM extraction unit or the third LSTM extraction unit;
each cell unit in the LSTM layer of the first LSTM extraction unit is used for inputting a marine pollution value at a moment;
each cell unit in the LSTM layer of the second LSTM extraction unit is used for inputting the change rate of the seaweed area at a moment;
each cell unit in the LSTM layer of the third LSTM extraction unit is configured to input a time-wise rate of change of the flocculent region.
Further, the expression of the salient feature enhancement layer is:
，
Wherein Y is the output feature of the salient feature enhancement layer, maxPool is the max-pooling operation, σ is the Sigmoid activation function, H is the input feature of the salient feature enhancement layer, Is multiplied by element.
The beneficial effects of the above further scheme are: the invention adopts the maximum pooling operation to extract the significant characteristic value output by the LSTM layer, then adopts the Sigmoid activation function to calculate the multiplication of the weight proportion of the significant characteristic value to the corresponding significant characteristic value, realizes the self-adaptive enhancement of the significant characteristic value, and adopts the average pooling layer to extract the global characteristic of each enhanced significant characteristic value.
Drawings
FIG. 1 is a flow chart of a marine pollution early warning method based on image recognition;
FIG. 2 is a schematic diagram of a parallel LSTM enhancement feature fusion neural network;
Fig. 3 is a schematic structural diagram of a first LSTM extraction unit, a second LSTM extraction unit, and a third LSTM extraction unit.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and all the inventions which make use of the inventive concept are protected by the spirit and scope of the present invention as defined and defined in the appended claims to those skilled in the art.
As shown in fig. 1, the marine pollution early warning method based on image recognition comprises the following steps:
S1, acquiring a submarine image in the ocean, and dividing the submarine image based on color smooth values of pixel points to obtain a seaweed area, a sea water area and a flocculent area;
s2, calculating a marine pollution value according to the color value of the seaweed area, the color value of the seawater area and the concentration value of the flocculent area;
s3, carrying out normalization processing on the ocean pollution value, the seaweed area change rate and the flocculent area change rate at each moment to obtain a sample;
s4, processing the sample by adopting a parallel LSTM enhanced feature fusion neural network to obtain a future ocean pollution value;
S5, early warning is carried out when the future pollution value of the ocean is larger than the pollution threshold value.
Seaweeds are an important component of the marine ecosystem, they form a wide grassland on the seafloor, providing habitat and food source for many marine organisms. However, when the sea is contaminated, the growth and reproduction of seaweed may be affected. Contaminants may reduce the photosynthetic efficiency of the seaweed, hamper its growth, and even lead to its death. In addition, some contaminants may alter the genes of seaweed, affecting the growth and viability of its offspring.
Healthy seaweed generally appears emerald and grows luxuriantly. When sea water is contaminated, the sea weed may be hindered in growth due to lack of illumination or nutrients, and the color may be dull or withered.
The color of seawater is an important indicator reflecting the condition of the ocean environment. Healthy seawater generally exhibits a clear blue or green color. And when pollution occurs, such as petroleum leakage or industrial wastewater discharge, the color of seawater may be changed to black, brown or other abnormal colors.
The step S1 comprises the following sub-steps:
S11, acquiring a submarine image in the ocean;
S12, denoising the underwater image to obtain an underwater denoising image;
S13, identifying a sea water area from the underwater denoising image according to the color smooth value of the pixel point;
s14, removing the sea water region from the underwater denoising image to obtain a plurality of independent closed subgraphs;
s15, identifying each independent closed subgraph to obtain a seaweed area and a flocculent area.
According to the method, after the underwater image in the ocean is acquired, denoising is carried out on the underwater image, the influence of noise points is removed, according to the color smoothing condition of each pixel point, a sea water area is firstly identified from the underwater denoising image, then the sea water area is removed from the underwater denoising image, and the rest area is segmented to obtain a sea weed area and a flocculent area.
In the present invention, capturing a water bottom image includes: the seaweed and the sea water are in two parts and flocculent distribution in the sea water.
The step S13 comprises the following substeps:
S131, calculating a color smooth value of each pixel point in the underwater denoising image;
s132, screening out pixel points with color smoothing values lower than a smoothing threshold value as candidate pixel points;
S133, clustering the candidate pixel points according to the color distance of each candidate pixel point to obtain a plurality of classifications;
S134, taking the classification with the largest number of candidate pixels as a sea water area.
In this embodiment, the color distance may be calculated by a difference between the pixel value of the candidate pixel and the pixel value of another candidate pixel.
The formula for calculating the color smoothing value in S131 is:
，
Wherein S is a color smoothing value, R o is an R channel value of a pixel, G o is a G channel value of the pixel, B o is a B channel value of the pixel, R o,i is an ith R channel value of a neighborhood range of the pixel, G o,i is an ith G channel value of the neighborhood range of the pixel, B o,i is an ith B channel value of the neighborhood range of the pixel, I is an absolute value, N is the number of R channel values, G channel values or B channel values of the neighborhood range, and i is a positive integer.
According to the method, the color smooth value of each pixel point is calculated according to the conditions of a plurality of R channel values, G channel values and B channel values in the neighborhood range of each pixel point, the smaller the color smooth value is, the more uniform the color is, the larger the color change in a seaweed area and a flocculent area is, meanwhile, after candidate pixel points are selected, clustering processing is carried out on the candidate pixel points according to the color distance of each candidate pixel point, the largest classification is selected as a sea water area, and the interference of smooth areas in part of the seaweed area and the flocculent area on the sea water area is avoided.
In this embodiment, the neighborhood range size is。
The step S15 comprises the following substeps:
S151, taking the independent closed subgraph with the largest area as a seaweed area;
S152, selecting an independent closed sub-graph with the area smaller than an area threshold from the rest multiple independent closed sub-graphs as a flocculent area.
The flocculent region is formed by a plurality of independent closed subgraphs.
In S14, the sea water area is removed from the underwater denoising image, so that the rest area to be segmented is formed by a plurality of independent closed subgraphs, in the ocean, seaweed and the seabed are connected together, the area ratio of the sea grass in the image is large, the independent closed subgraphs with the largest area are used as the seaweed area, and the independent closed subgraphs with the area smaller than the area threshold are selected from the rest independent closed subgraphs, so that floccules distributed in the sea water are screened out, the independent closed subgraphs with the area larger than the area threshold are discarded, and the interference of fish is avoided.
The calculation formula of the concentration value of the flocculent region in the S2 is as follows:
，
Wherein μ is a concentration value of a flocculent region, E F is an area of the flocculent region, and E S is an area of a seawater region;
the formula for calculating the ocean pollution value in the S2 is as follows:
，
wherein, gamma is the ocean pollution value, C G is the color value of the seaweed area, C S is the color value of the sea water area and is the color standard value of the sea grass area,Is the standard value of the color of the sea water area,As a standard value of the concentration of the flocculent region, alpha 1 is a first weighting coefficient, alpha 2 is a second weighting coefficient, and alpha 3 is a third weighting coefficient.
In this embodiment, the color standard value of the sea weed area, the color standard value of the sea water area, and the flocculent area concentration standard value are set values according to the ocean environment or values stored when the ocean is not polluted.
In this embodiment, the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient are weighting values set according to requirements or experience.
The invention synthesizes the difference between the color value of the sea weed area and the color standard value of the sea weed area, the difference between the color value of the sea water area and the color standard value of the sea water area, the difference between the concentration value of the flocculent area and the concentration standard value of the flocculent area, and obtains the ocean pollution value through weighting treatment, and whether the ocean is abnormal or not is reflected from three aspects.
The calculation formula of the change rate of the seaweed area in the S3 is as follows:
，
Wherein V G,t is the change rate of the seaweed area at the T moment, E G,t is the area of the seaweed area at the T moment, E G,t-1 is the area of the seaweed area at the T-1 moment, and T is the interval time length;
the calculation formula of the flocculent region change rate in the S3 is as follows:
，
Wherein V F,t is the change rate of the flocculent region at the t time, E F,t is the flocculent region area at the t time, E F,t-1 is the flocculent region area at the t-1 time, and t is the number of the time.
The invention calculates the change rate of the seaweed area and the change rate of the flocculent area, reflects whether the seaweed area has rapid change in a short time through the change rate of the seaweed area, reflects whether the flocculent area has rapid change in a short time through the change rate of the flocculent area, and can show that the environment in the ocean changes rapidly when the seaweed area and the flocculent area change rapidly.
As shown in fig. 2, the parallel LSTM enhancement feature fusion neural network in S4 includes: the device comprises a first LSTM extraction unit, a second LSTM extraction unit, a third LSTM extraction unit, concat layers and a full-connection layer;
the input end of the first LSTM extraction unit is used for inputting ocean pollution values at all moments;
the input end of the second LSTM extraction unit is used for inputting the change rate of the seaweed area at each moment;
The input end of the third LSTM extraction unit is used for inputting the flocculent region change rate at each moment;
The input end of the Concat layers is respectively connected with the output end of the first LSTM extraction unit, the output end of the second LSTM extraction unit and the output end of the third LSTM extraction unit; the input end of the full-connection layer is connected with the output end of the Concat layers, and the output end of the full-connection layer is used as the output end of the parallel LSTM enhancement feature fusion neural network.
The invention inputs ocean pollution values at all times at a first LSTM extraction unit, captures the association condition of the ocean pollution values at all times, inputs the seaweed area change rate at all times at a second LSTM extraction unit, captures the association condition of the seaweed area change rate at all times, inputs the flocculent area change rate at all times at a third LSTM extraction unit, captures the association condition of the flocculent area change rate at all times, splices the output characteristics of three LSTM extraction units through Concat layers, and outputs the ocean future pollution values by adopting a full connection layer.
As shown in fig. 3, the first LSTM extraction unit, the second LSTM extraction unit, and the third LSTM extraction unit each include: LSTM layer, salient feature enhancement layer and average pooling layer;
The input end of the LSTM layer is used as the input end of the first LSTM extraction unit, the second LSTM extraction unit or the third LSTM extraction unit, and the output end of the LSTM layer is connected with the input end of the salient feature enhancement layer; the input end of the average pooling layer is connected with the output end of the salient feature enhancement layer, and the output end of the average pooling layer is used as the output end of the first LSTM extraction unit, the second LSTM extraction unit or the third LSTM extraction unit;
each cell unit in the LSTM layer of the first LSTM extraction unit is used for inputting a marine pollution value at a moment;
each cell unit in the LSTM layer of the second LSTM extraction unit is used for inputting the change rate of the seaweed area at a moment;
each cell unit in the LSTM layer of the third LSTM extraction unit is configured to input a time-wise rate of change of the flocculent region.
The expression of the salient feature enhancement layer is:
，
Wherein Y is the output feature of the salient feature enhancement layer, maxPool is the max-pooling operation, σ is the Sigmoid activation function, H is the input feature of the salient feature enhancement layer, Is multiplied by element.
The invention adopts the maximum pooling operation to extract the significant characteristic value output by the LSTM layer, adopts the Sigmoid activation function to calculate the weight proportion of the significant characteristic value, multiplies the weight proportion with the corresponding significant characteristic value to realize the self-adaptive enhancement of the significant characteristic value, and adopts the average pooling layer to extract the global characteristic of each enhanced significant characteristic value.
In this embodiment, all thresholds are set empirically or experimentally.
According to the method, a marine pollution value is calculated according to the color value of the sea grass area, the color value of the sea water area and the concentration value of the flocculent area, so that analysis of the current marine pollution condition is realized, a parallel LSTM enhanced characteristic fusion neural network is adopted to process samples, the marine future pollution value is predicted by comprehensively combining the marine pollution value, the sea grass area change rate and the flocculent area change rate at each moment, the marine pollution change condition is embodied through the marine pollution value at each moment, the sea grass change condition is embodied through the sea grass area change rate, the flocculent area change rate is embodied by the flocculent change condition, the characteristic of each moment change is extracted by adopting the parallel LSTM enhanced characteristic fusion neural network, the precision of predicting the marine future pollution value is improved, and the marine pollution is early-warned.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (7)
1. The marine pollution early warning method based on image recognition is characterized by comprising the following steps of:
S1, acquiring a submarine image in the ocean, and dividing the submarine image based on color smooth values of pixel points to obtain a seaweed area, a sea water area and a flocculent area;
s2, calculating a marine pollution value according to the color value of the seaweed area, the color value of the seawater area and the concentration value of the flocculent area;
s3, carrying out normalization processing on the ocean pollution value, the seaweed area change rate and the flocculent area change rate at each moment to obtain a sample;
s4, processing the sample by adopting a parallel LSTM enhanced feature fusion neural network to obtain a future ocean pollution value;
s5, early warning is carried out when the future pollution value of the ocean is larger than a pollution threshold value;
the parallel LSTM enhancement feature fusion neural network in the S4 comprises: the device comprises a first LSTM extraction unit, a second LSTM extraction unit, a third LSTM extraction unit, concat layers and a full-connection layer;
the input end of the first LSTM extraction unit is used for inputting ocean pollution values at all moments;
the input end of the second LSTM extraction unit is used for inputting the change rate of the seaweed area at each moment;
The input end of the third LSTM extraction unit is used for inputting the flocculent region change rate at each moment;
The input end of the Concat layers is respectively connected with the output end of the first LSTM extraction unit, the output end of the second LSTM extraction unit and the output end of the third LSTM extraction unit; the input end of the full-connection layer is connected with the output end of the Concat layers, and the output end of the full-connection layer is used as the output end of the parallel LSTM enhancement feature fusion neural network;
the first LSTM extraction unit, the second LSTM extraction unit, and the third LSTM extraction unit each include: LSTM layer, salient feature enhancement layer and average pooling layer;
The input end of the LSTM layer is used as the input end of the first LSTM extraction unit, the second LSTM extraction unit or the third LSTM extraction unit, and the output end of the LSTM layer is connected with the input end of the salient feature enhancement layer; the input end of the average pooling layer is connected with the output end of the salient feature enhancement layer, and the output end of the average pooling layer is used as the output end of the first LSTM extraction unit, the second LSTM extraction unit or the third LSTM extraction unit;
each cell unit in the LSTM layer of the first LSTM extraction unit is used for inputting a marine pollution value at a moment;
each cell unit in the LSTM layer of the second LSTM extraction unit is used for inputting the change rate of the seaweed area at a moment;
each cell unit in the LSTM layer of the third LSTM extraction unit is used for inputting the flocculent region change rate at a moment;
the expression of the salient feature enhancement layer is:
，
Wherein Y is the output feature of the salient feature enhancement layer, maxPool is the max-pooling operation, σ is the Sigmoid activation function, H is the input feature of the salient feature enhancement layer, Is multiplied by element.
2. The marine pollution early warning method based on image recognition according to claim 1, wherein the S1 comprises the following sub-steps:
S11, acquiring a submarine image in the ocean;
S12, denoising the underwater image to obtain an underwater denoising image;
S13, identifying a sea water area from the underwater denoising image according to the color smooth value of the pixel point;
s14, removing the sea water region from the underwater denoising image to obtain a plurality of independent closed subgraphs;
s15, identifying each independent closed subgraph to obtain a seaweed area and a flocculent area.
3. The marine pollution early warning method based on image recognition according to claim 2, wherein the S13 comprises the following sub-steps:
S131, calculating a color smooth value of each pixel point in the underwater denoising image;
s132, screening out pixel points with color smoothing values lower than a smoothing threshold value as candidate pixel points;
S133, clustering the candidate pixel points according to the color distance of each candidate pixel point to obtain a plurality of classifications;
S134, taking the classification with the largest number of candidate pixels as a sea water area.
4. The marine pollution early warning method based on image recognition according to claim 3, wherein the formula for calculating the color smoothing value in S131 is:
，
Wherein S is a color smoothing value, R o is an R channel value of a pixel, G o is a G channel value of the pixel, B o is a B channel value of the pixel, R o,i is an ith R channel value of a neighborhood range of the pixel, G o,i is an ith G channel value of the neighborhood range of the pixel, B o,i is an ith B channel value of the neighborhood range of the pixel, I is an absolute value, N is the number of R channel values, G channel values or B channel values of the neighborhood range, and i is a positive integer.
5. The marine pollution early warning method based on image recognition according to claim 2, wherein the step S15 comprises the following sub-steps:
S151, taking the independent closed subgraph with the largest area as a seaweed area;
S152, selecting an independent closed sub-graph with the area smaller than an area threshold from the rest multiple independent closed sub-graphs as a flocculent area.
6. The marine pollution early warning method based on image recognition according to claim 1, wherein the calculation formula of the concentration value of the flocculent region in S2 is:
，
Wherein μ is a concentration value of a flocculent region, E F is an area of the flocculent region, and E S is an area of a seawater region;
the formula for calculating the ocean pollution value in the S2 is as follows:
，
wherein, gamma is the ocean pollution value, C G is the color value of the seaweed area, C S is the color value of the sea water area and is the color standard value of the sea grass area,Is the standard value of the color of the sea water area,As a standard value of the concentration of the flocculent region, alpha 1 is a first weighting coefficient, alpha 2 is a second weighting coefficient, and alpha 3 is a third weighting coefficient.
7. The marine pollution early warning method based on image recognition according to claim 1, wherein the calculation formula of the seaweed area change rate in S3 is:
，
Wherein V G,t is the change rate of the seaweed area at the T moment, E G,t is the area of the seaweed area at the T moment, E G,t-1 is the area of the seaweed area at the T-1 moment, and T is the interval time length;
the calculation formula of the flocculent region change rate in the S3 is as follows:
，
Wherein V F,t is the change rate of the flocculent region at the t time, E F,t is the flocculent region area at the t time, E F,t-1 is the flocculent region area at the t-1 time, and t is the number of the time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410790912.7A CN118379611B (en) | 2024-06-19 | 2024-06-19 | Ocean pollution early warning method based on image recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410790912.7A CN118379611B (en) | 2024-06-19 | 2024-06-19 | Ocean pollution early warning method based on image recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118379611A CN118379611A (en) | 2024-07-23 |
| CN118379611B true CN118379611B (en) | 2024-08-23 |
Family
ID=91904196
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410790912.7A Active CN118379611B (en) | 2024-06-19 | 2024-06-19 | Ocean pollution early warning method based on image recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118379611B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117611588A (en) * | 2024-01-23 | 2024-02-27 | 浙江大学海南研究院 | Marine environment pollution early detection and early warning method |
| CN117951584A (en) * | 2024-03-13 | 2024-04-30 | 青岛启弘信息科技有限公司 | Ocean data processing and information scheduling system based on AI and Internet of things technology |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8454838B2 (en) * | 2011-03-30 | 2013-06-04 | Crystal Lagoons (Curacao) B.V. | Method and system for the sustainable cooling of industrial processes |
| KR101735198B1 (en) * | 2015-09-02 | 2017-05-12 | 경남과학기술대학교 산학협력단 | Measurement Apparatus of Marine Environment Pollution Indicator using a Surface Roughness of Macrophthalmus Japonicus |
| CN106940363B (en) * | 2017-03-14 | 2019-04-30 | 山东省科学院海洋仪器仪表研究所 | A kind of marine pollution method for early warning based on marine organisms behavior reaction |
-
2024
- 2024-06-19 CN CN202410790912.7A patent/CN118379611B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117611588A (en) * | 2024-01-23 | 2024-02-27 | 浙江大学海南研究院 | Marine environment pollution early detection and early warning method |
| CN117951584A (en) * | 2024-03-13 | 2024-04-30 | 青岛启弘信息科技有限公司 | Ocean data processing and information scheduling system based on AI and Internet of things technology |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118379611A (en) | 2024-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Marini et al. | Tracking fish abundance by underwater image recognition | |
| Ma et al. | A state-of-the-art survey of object detection techniques in microorganism image analysis: from classical methods to deep learning approaches | |
| Kulwa et al. | A state-of-the-art survey for microorganism image segmentation methods and future potential | |
| Spampinato et al. | Detecting, tracking and counting fish in low quality unconstrained underwater videos | |
| Beyenal et al. | Quantifying biofilm structure: facts and fiction | |
| Pennekamp et al. | Implementing image analysis in laboratory‐based experimental systems for ecology and evolution: a hands‐on guide | |
| Bi et al. | A semi-automated image analysis procedure for in situ plankton imaging systems | |
| Marini et al. | Automated estimate of fish abundance through the autonomous imaging device GUARD1 | |
| Yu et al. | Floating pollutant image target extraction algorithm based on immune extremum region | |
| Zhao et al. | An adaptive method of damage detection for fishing nets based on image processing technology | |
| Elawady | Sparse coral classification using deep convolutional neural networks | |
| Li et al. | Detection of uneaten fish food pellets in underwater images for aquaculture | |
| Cheng et al. | A baseline for microplastic particle occurrence and distribution in Great Bay Estuary | |
| Joo et al. | Identification of cichlid fishes from Lake Malawi using computer vision | |
| Zhang et al. | A comprehensive survey with quantitative comparison of image analysis methods for microorganism biovolume measurements | |
| CN109389613B (en) | Residual bait counting method based on computer vision | |
| CN112001641B (en) | Scallop breeding area suitability remote sensing evaluation system | |
| CN112464742B (en) | Method and device for automatically identifying red tide image | |
| CN118379611B (en) | Ocean pollution early warning method based on image recognition | |
| CN114005064A (en) | Biological water body pollution early warning method and device based on machine vision technology | |
| Ovchinnikova et al. | Exploring the potential to use low cost imaging and an open source convolutional neural network detector to support stock assessment of the king scallop (Pecten maximus) | |
| Carloto et al. | Detection of morphological changes caused by chemical stress in the cyanobacterium Planktothrix agardhii using convolutional neural networks | |
| CN115131657A (en) | Marine ecology analysis method and system | |
| Ranolo et al. | Underwater and coastal seaweeds detection for fluorescence seaweed photos and videos using YOLOV3 and YOLOV5 | |
| Dannemiller et al. | A new method for the segmentation of algae images using retinex and support vector machine |
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 |