CN117455743B

CN117455743B - Deep learning-based dredging mud resource utilization evaluation method and system

Info

Publication number: CN117455743B
Application number: CN202311740876.5A
Authority: CN
Inventors: 王琰; 严金辉; 关瑶; 薛飞; 杨帆; 石军; 王超; 石萍; 罗艳; 黄毅; 王亮
Original assignee: State Ocean Administration South China Sea Planning And Environment Research Institute
Current assignee: State Ocean Administration South China Sea Planning And Environment Research Institute
Priority date: 2023-12-18
Filing date: 2023-12-18
Publication date: 2024-04-09
Anticipated expiration: 2043-12-18
Also published as: CN117455743A

Abstract

The invention provides a dredging mud resource utilization evaluation method and system based on deep learning, and belongs to the technical field of resource utilization. Firstly, a dredging mud sample is obtained; secondly, carrying out physical analysis, chemical analysis and biological analysis on the dredging mud sample to obtain physical characteristics, chemical characteristics and biological characteristics; inputting the dredging mud sample image into an image feature extraction network for feature extraction to obtain image features; inputting the image features, the physical features, the chemical features and the biological features into a resource network for identification, and outputting a resource utilization mode; and finally, planning according to the resource utilization mode to obtain a planning scheme. The invention can effectively evaluate the recycling potential of the dredging mud, improve the treatment efficiency and the recycling level of the dredging mud, reduce the environmental influence and the treatment cost of the dredging mud, and realize the efficient and accurate evaluation of the recycling potential of the dredging mud.

Description

Deep learning-based dredging mud resource utilization evaluation method and system

Technical Field

The invention belongs to the technical field of resource utilization, and particularly relates to a dredging mud resource utilization evaluation method and system based on deep learning.

Background

Dredging mud is a sediment mixture with higher water content, which is generated when dredging in water areas such as rivers, lakes, ports, channels and the like, and is a typical engineering waste. The production amount of the dredging mud is huge, and the dredging mud produced every year in China is about 10 hundred million cubic meters and accounts for 1/4 of the total world. The treatment and disposal of the dredging mud is a serious environmental problem, and because the dredging mud contains a large amount of pollutants such as organic matters, heavy metals, nutrient salts and the like, if the dredging mud is not treated, the dredging mud is directly discharged or buried, and serious consequences such as water eutrophication, soil pollution, ecological damage and the like can be caused. Therefore, the treatment and disposal of the dredging mud should follow the principles of reduction, harmlessness and recycling, and convert the dredging mud into valuable resources as much as possible, thereby realizing the recycling and green development of the dredging mud. Traditional dredging mud analysis methods focus on a single aspect of physical, chemical or biological characteristics, and cannot fully evaluate the recycling potential thereof. Furthermore, these methods are generally time consuming and require high sample handling and analysis. Therefore, it is necessary to develop a dredging mud resource utilization evaluation method based on deep learning.

Disclosure of Invention

Based on the technical problems, the invention provides a dredging mud recycling evaluation method and system based on deep learning, which analyze and predict the characteristics and the utilization mode of the dredging mud by using a deep learning technology and improve the recycling efficiency and effect of the dredging mud.

The invention provides a dredging mud resource utilization evaluation method based on deep learning, which comprises the following steps:

step S1: acquiring a dredging mud sample;

step S2: performing physical analysis, chemical analysis and biological analysis on the dredging mud sample to obtain physical characteristics, chemical characteristics and biological characteristics;

step S3: inputting the dredging mud sample image into an image feature extraction network to perform feature extraction to obtain image features;

step S4: inputting the image features, the physical features, the chemical features and the biological features into a resource network for identification, and outputting a resource utilization mode;

step S5: planning is carried out according to the resource utilization mode, and a planning scheme is obtained.

Optionally, the physical analysis, chemical analysis and biological analysis are performed on the dredged mud sample to obtain physical characteristics, chemical characteristics and biological characteristics, which specifically include:

The formula of the water content of the dredging mud is as follows:

in the method, in the process of the invention,is the water content of the dredging mud; />Is the mass of water; />Is solid mass;

the formula of the average particle diameter of the dredging mud is as follows:

in the method, in the process of the invention,is the average particle diameter of the dredging mud; />Is the first/>Particle diameter; />Is->The individual particle frequencies; />The number of the particles;

the formula of the temperature influence coefficient of the dredging mud is as follows:

in the method, in the process of the invention,is the temperature influence coefficient of the dredging mud; />Viscosity of the dredged mud; />Is the reference viscosity; />Is the activation energy;is a gas constant; />Is the temperature of the dredging mud; />Is the reference temperature;

in the method, in the process of the invention,van der Waals forces between dredging mud particles; />Is a hamilton constant; />Is the thickness of the dredging mud; />Is the radius of the particles; />Is the inter-particle distance;

the formula of the dredged mud porosity is as follows:

in the method, in the process of the invention,is the dredged mud porosity; />Is the total volume of the dredging mud; />Is the solid volume of dredging mud;

the Young modulus formula of the dredging mud is as follows:

in the method, in the process of the invention,young's modulus for dredging mud particles; />Stress of the particles; />Strain for the particles;

the dredging mud source and the environmental impact coefficient formula are as follows:

in the method, in the process of the invention,is the dredging mud source and the environmental impact coefficient; />Is->The content of the seed compound; />Is->Weights of seed compounds; / >Number of species of the compound;

the formula of the dredging mud structure and the morphological influence coefficient is as follows:

in the method, in the process of the invention,is the structure and form influence coefficient of the dredging mud; />Is->The content of the seed compound; />Is->A structural or morphological score of the seed compound; />The number of kinds of the compounds;

the formula of the dredging mud time and space influence coefficient is as follows:

in the method, in the process of the invention,is the dredging mud time and space influence coefficient; />Is->The content of the seed compound; />Is->Time score of seed compound; />Is->Spatial score of the seed compound; />The number of kinds of the compounds;

the dredging mud pollution evaluation index formula is:

in the method, in the process of the invention,evaluating an index for dredging mud pollution; />Is->Seed pollutant content; />Is->Toxicity coefficient of seed pollutant; />The number of contaminant species;

the formula of migration and conversion coefficients of the dredging mud is as follows:

in the method, in the process of the invention,migration and conversion coefficients for the dredged mud; />Is->Seed pollutant content; />Is->Seed contaminant migration rate; />Is->Rate of conversion of the seed contaminant; />The number of contaminant species;

the formula of the microorganism number of the dredging mud is as follows:

in the method, in the process of the invention,the microbial count of the dredging mud; />Is the initial microbial population; />Is the microorganism growth rate; />Time is;

The formula of the microorganism metabolic activity of the dredging mud is as follows:

in the method, in the process of the invention,microbial metabolic activity for dredging mud; />A change in concentration of a substrate or product that is consumed or produced by the microorganism over a period of time; />Is a time interval.

Optionally, the inputting the dredging mud sample image into an image feature extraction network for feature extraction to obtain image features specifically includes:

sequentially inputting dredging mud sample images into a first standard convolution layer, a first standardization activation layer and a first maximum pooling layer to carry out rolling and pooling operation, and obtaining a characteristic diagram S3;

inputting the feature map S3 into a first structured dynamic convolution module, a second normalized activation layer, a second structured dynamic convolution module and a third normalized activation layer in sequence to perform dynamic convolution operation to obtain a feature map S7;

inputting the feature map S7 to a ternary attention module and a multi-scale recursive attention module in sequence to perform attention operation, so as to obtain a feature map S9;

sequentially inputting the feature map S9 into a second standard convolution layer, a fourth normalized activation layer, a third standard convolution layer, a first normalization layer, a fourth standard convolution layer, a second normalization layer and an element addition layer to carry out convolution and element addition operation to obtain a feature map S16;

And sequentially inputting the feature map S16 into a first activation function layer, a second maximum pooling layer, a fifth standard convolution layer, a third maximum pooling layer and a first global average pooling layer to perform feature extraction operation, so as to obtain image features S21.

Optionally, the first structured dynamic convolution module specifically includes:

calculating the average value of the input tensor in the height dimension and the width dimension to obtain dynamic input data;

generating the weight of the dynamic input data by using a full connection layer to obtain a dynamic convolution kernel weight;

remolding the dynamic convolution kernel weight to obtain a dynamic convolution kernel;

and carrying out convolution operation on the input tensor by using a dynamic convolution check to obtain an output tensor.

Optionally, the multi-scale recursive attention module specifically includes:

initializing an attention map list and setting a target size;

traversing a scale list, creating a corresponding number of ternary attention modules according to the number of scales in the scale list, and operating an input feature map of each scale, wherein the method specifically comprises the following steps:

carrying out average pooling on the input feature map by using an average pooling layer; repeatedly applying a ternary attention module to the pooled feature map; adjusting the processed feature map to a target size; adding the resized feature graphs to the attention profile list; combining attention diagrams on all scales to obtain a final output characteristic diagram.

The invention also provides a dredging mud resource utilization evaluation system based on deep learning, which comprises:

the data sample acquisition module is used for acquiring a dredging mud sample;

the basic characteristic extraction module is used for carrying out physical analysis, chemical analysis and biological analysis on the dredging mud sample to obtain physical characteristics, chemical characteristics and biological characteristics;

the image feature extraction module is used for inputting the dredging mud sample image into the image feature extraction network to perform feature extraction to obtain image features;

the resource utilization classification module is used for inputting the image features, the physical features, the chemical features and the biological features into a resource network for identification and outputting a resource utilization mode;

and the standard scheme making module is used for planning according to the resource utilization mode to obtain a planning scheme.

Optionally, the basic characteristic extraction module specifically includes:

the formula of the water content of the dredging mud is as follows:

the formula of the average particle diameter of the dredging mud is as follows:

in the method, in the process of the invention,is the average particle diameter of the dredging mud; />Is->Particle diameter; />Is->The individual particle frequencies; / >The number of the particles;

the formula of the dredged mud porosity is as follows:

in the method, in the process of the invention,is the dredged mud porosity; />Is the total volume of the dredging mud; />Is a dredged mud solidA volume;

the Young modulus formula of the dredging mud is as follows:

in the method, in the process of the invention,is the dredging mud source and the environmental impact coefficient; />Is->The content of the seed compound; />Is->Weights of seed compounds; />Number of species of the compound;

the dredging mud pollution evaluation index formula is:

the formula of the microorganism number of the dredging mud is as follows:

Optionally, the image feature extraction module specifically includes:

The first standard convolution module is used for inputting dredging mud sample images into the first standard convolution layer, the first normalized activation layer and the first maximum pooling layer in sequence to carry out convolution and pooling operations to obtain a characteristic diagram S3;

the structured dynamic convolution module is used for inputting the feature map S3 into the first structured dynamic convolution module, the second normalized activation layer, the second structured dynamic convolution module and the third normalized activation layer in sequence to perform dynamic convolution operation to obtain a feature map S7;

the attention mechanism module is used for sequentially inputting the feature map S7 into the ternary attention module and the multi-scale recursive attention module to perform attention operation to obtain a feature map S9;

the residual error module is used for inputting the characteristic diagram S9 into a second standard convolution layer, a fourth normalized activation layer, a third standard convolution layer, a first normalization layer, a fourth standard convolution layer, a second normalization layer and an element addition layer in sequence to carry out convolution and element addition operation to obtain a characteristic diagram S16;

and the image feature output module is used for sequentially inputting the feature map S16 into the first activation function layer, the second maximum pooling layer, the fifth standard convolution layer, the third maximum pooling layer and the first global average pooling layer to perform feature extraction operation, so as to obtain image features S21.

the input sub-module is used for calculating the average value of the input tensor in the height dimension and the width dimension to obtain dynamic input data;

the dynamic convolution kernel weight generation sub-module is used for generating the weight of the dynamic input data by using the full connection layer to obtain the dynamic convolution kernel weight;

the dynamic convolution kernel acquisition submodule is used for remolding the dynamic convolution kernel weight to obtain a dynamic convolution kernel;

and the output sub-module is used for carrying out convolution operation on the input tensor by using a dynamic convolution check to obtain an output tensor.

Optionally, the multi-scale recursive attention module specifically includes:

an initialization sub-module for initializing an attention map list and setting a target size;

the multi-scale operation sub-module is used for traversing the scale list, creating a corresponding number of ternary attention modules according to the scale number in the scale list, and operating the input feature map of each scale, and specifically comprises the following steps:

Compared with the prior art, the invention has the following beneficial effects:

the invention can comprehensively analyze the physical, chemical, biological and image characteristics of the dredging mud by utilizing the deep learning technology, extract key characteristic parameters of the dredging mud and provide data support for the resource utilization evaluation of the dredging mud; the method comprises the steps of intelligently identifying the resource utilization mode of the dredging mud, outputting the optimal resource utilization mode of the dredging mud, and providing decision support for resource utilization planning of the dredging mud; effectively improves the recycling level of the dredging mud, reduces the treatment cost and the environmental impact of the dredging mud, and realizes the recycling and green development of the dredging mud.

Drawings

FIG. 1 is a flow chart of a dredging mud recycling evaluation method based on deep learning;

FIG. 2 is a diagram of an image feature extraction network according to the present invention;

FIG. 3 is a diagram of a resource network architecture of the present invention;

fig. 4 is a block diagram of a dredging mud recycling evaluation system based on deep learning.

Detailed Description

The invention is further described below in connection with specific embodiments and the accompanying drawings, but the invention is not limited to these embodiments.

Example 1

As shown in fig. 1, the invention discloses a dredging mud recycling evaluation method based on deep learning, which comprises the following steps:

Step S1: a dredged sediment sample is obtained.

Step S2: and carrying out physical analysis, chemical analysis and biological analysis on the dredging mud sample to obtain physical characteristics, chemical characteristics and biological characteristics.

Step S3: inputting the dredging mud sample image into an image feature extraction network to perform feature extraction to obtain image features.

Step S4: the image features, the physical features, the chemical features and the biological features are input into a resource network for identification, and a resource utilization mode is output.

The steps are discussed in detail below:

step S1: a dredged sediment sample is obtained.

The step S1 specifically comprises the following steps:

in this example, the generation of dredged mud exists in several ways: dredging activities may be periodic, such as periodic cleaning of a river, port or channel; in certain engineering projects, such as the construction of new ports or reservoirs, dredged mud can be intensively produced during the engineering. In certain large, long-term dredging projects, dredged mud may continue to be produced. With respect to acquiring dredged mud samples, if the dredging activities are periodic or project-wise, the data may be collected in batches, after each dredging activity, a batch of dredged mud samples may be collected for analysis; for continuous dredging projects, continuous data monitoring and acquisition may need to be implemented; this may include periodic (e.g., weekly, monthly) sampling or continuous collection of data using automated monitoring equipment; in some high-tech projects, data on the dredged mud may be collected in real-time using sensors and on-line monitoring equipment.

The generation and treatment of dredged mud is a continuous process, typically by batch-wise cleaning, transporting and disposing of the dredged mud according to the division of the dredging area and depth. The quantity and quality of the dredged mud may vary from batch to batch, and therefore need to be adjusted and optimized for the actual situation.

The acquisition data of the dredging mud refers to data for detecting and analyzing physical, chemical and biological characteristics of the dredging mud, which are important for evaluating the pollution degree and the treatment effect of the dredging mud. The sediment is sampled by adopting different sampling and detection methods, such as drilling, grab bucket or underwater robot, and the sediment is detected by adopting a site rapid detection instrument or laboratory analysis instrument.

The step S2 specifically comprises the following steps:

physical analysis of a dredged sediment sample, specifically comprising:

the fluidity of the dredging mud is related to the water content, the higher the water content is, the stronger the fluidity is, and the formula is:

in the method, in the process of the invention,the water content of the dredged mud; />Is the mass of water; / >Is the mass of the solid. The parameter acquisition process comprises the following steps: putting a certain amount of dredging mud into an oven, heating until the water is evaporated, and weighing the dry weight of the dredging mud; the dried dredging mud is put into water and fully stirred, so that the original water content is restored, and the wet weight of the dredging mud is weighed; calculating the water content of the dredging mud according to a formula. The formula is applied to actual work, and the fluidity of the dredging mud can be regulated by measuring and controlling the water content of the dredging mud, so that the formula is suitable for different recycling modes. For example, if the dredging mud is to be used for land reclamation, the water content of the dredging mud can be reduced, and the stability of the dredging mud can be improved; if the dredging mud is used for ecological restoration, the water content of the dredging mud can be increased, and the permeability of the dredging mud can be increased.

The fluidity of the dredged mud is also related to its particle size, the finer the particles, the stronger the fluidity, the formula:

in the method, in the process of the invention,is the average particle diameter of the dredging mud; />Is->Particle diameter; />Is->The frequency of the individual particles; />Is the number of particles. The parameter acquisition process comprises the following steps: carrying out granularity analysis on a certain amount of dredging mud to obtain frequency distribution of particles with different sizes; each particle was observed and its diameter measured; calculating the average particle diameter of the dredging mud according to the formula +. >. The formula is applied to actual work, and the fluidity of the dredging mud can be regulated by measuring and controlling the average particle diameter of the dredging mud, so that the formula is suitable for different recycling modes. For example, if the dredging mud is to be used in building material preparation, the average particle diameter of the dredging mud can be increased, and the strength thereof can be improved; if the dredging mud is to be used in other ways, the average particle diameter of the dredging mud can be reduced, increasing its uniformity.

The fluidity of the dredged mud is also related to the temperature thereof, the higher the temperature is, the stronger the fluidity is, the formula is:

in the method, in the process of the invention,is the temperature influence coefficient of the dredging mud; />Viscosity of the dredged mud; />Is the reference viscosity; />Is the activation energy;is a gas constant; />Is dredgingThe temperature of the mud; />Is the reference temperature. The parameter acquisition process comprises the following steps: putting a certain amount of dredging mud into a constant-temperature water bath, and adjusting the temperature of the water bath to enable the dredging mud to reach different temperatures; carrying out viscosity test on the dredging mud at each temperature to obtain a viscosity value of the dredging mud; calculating the influence coefficient of the temperature of the dredging mud on the fluidity thereof according to the formula>. The formula is applied to actual work, and the fluidity of the dredging mud can be regulated by measuring and controlling the temperature of the dredging mud, so that the formula is suitable for different recycling modes. For example, if the dredging mud is to be used for ecological restoration, the temperature of the dredging mud can be reduced, and the influence of the dredging mud on the temperature of the water body can be reduced; if the dredging mud is to be used in other ways, the temperature of the dredging mud can be increased, the viscosity of the dredging mud can be reduced, and the fluidity of the dredging mud can be increased.

The stability of the dredged mud is related to its inter-particle interactions, the stronger the inter-particle interactions, the higher the stability, the formula:

in the method, in the process of the invention,van der Waals forces between dredging mud particles; />Is a hamilton constant; />Is the thickness of the dredging mud; />Is the radius of the particles; />Is the distance between the particles. This isThe parameter acquisition process comprises the following steps: carrying out X-ray diffraction analysis on a certain amount of dredging mud to obtain structural and morphological information of particles, such as lattice constant, interplanar spacing and the like; according to the structure and morphology information of the particles, calculating the Hamiltonian constant>This is a constant that is related to the physical properties of the particles; carrying out a compression test on a certain amount of dredging mud to obtain a positive pressure among particles, wherein the positive pressure is related to the distance; calculating Van der Waals force between particles of dredging mud according to formula>. The formula is applied in actual work, and the stability of the dredging mud can be regulated by measuring and controlling the Van der Waals force among particles of the dredging mud, so that the dredging mud is suitable for different recycling modes. For example, if the dredging mud is to be used for sea-filling and land-making, the van der Waals forces between the particles of the dredging mud can be increased, improving its stability; if the dredged mud is to be used in other ways, the inter-particle van der Waals forces of the dredged mud can be reduced, reducing its cohesiveness.

The stability of the dredged mud is also related to the alignment of the particles, the tighter the particles, the higher the stability, the formula:

in the method, in the process of the invention,porosity of the dredged mud; />Is the total volume of dredging mud; />Is the solid volume of the dredged mud. The parameter acquisition process comprises the following steps: putting a certain amount of dredging mud into a densitometer, measuring its total volume +.>The method comprises the steps of carrying out a first treatment on the surface of the Taking out dredging mud from densitometer, placing into oven, heating until water is evaporated, and weighing dry weight +.>The method comprises the steps of carrying out a first treatment on the surface of the According to the dry weight of the dredging mud->And Density->Calculate its solid volume +.>The method comprises the steps of carrying out a first treatment on the surface of the Calculating the porosity of the particles of the dredging mud according to the formula>. The formula is applied to actual work, and the stability of the dredging mud can be adjusted by measuring and controlling the porosity of the particles of the dredging mud, so that the formula is suitable for different recycling modes. For example, if the dredging mud is to be used for ecological restoration, the porosity of the particles of the dredging mud can be increased, increasing its permeability; if the dredged mud is to be used in other ways, the porosity of the particles of the dredged mud can be reduced, increasing its compactness.

The stability of a dredged mud is also related to the elasticity of its particles, the more elastic the particles are, the higher the stability, the formula:

in the method, in the process of the invention, Young's modulus for dredging mud particles; />Stress of the particles; />Is in the form of granuleStrain of the pellet. The parameter acquisition process comprises the following steps: subjecting a quantity of dredging mud to compression test to obtain stress of its particles>And Strain->Is a relationship of (2); fitting a curve in the elastic region by least squares to obtain a slope +.>I.e. the young's modulus of the particles of the dredging mud. The formula is applied in actual work, and the stability of the dredging mud can be adjusted by measuring and controlling the Young modulus of the particles of the dredging mud, so that the dredging mud is suitable for different recycling modes. For example, if the dredging mud is to be used in building material preparation, the Young's modulus of the particles of the dredging mud can be increased, improving the strength thereof; if the dredging mud is to be used in other ways, the Young's modulus of the particles of the dredging mud can be reduced, increasing its plasticity.

Carrying out chemical analysis on the dredged mud sample, wherein the chemical analysis specifically comprises the following steps:

the chemical composition of dredging mud is related to its source and environment, which can lead to different chemical compositions, the formula being:

in the method, in the process of the invention,the coefficient of influence of the source of the dredging mud and the environment on the chemical components of the dredging mud; />Is->The content of the seed chemical element or compound; / >Is->The weight of the seed chemical element or compound; />(Chemical/Compounds Types) is the number of Types of Chemical elements or Compounds. The parameter acquisition process comprises the following steps: subjecting a quantity of dredging mud to chemical analysis to obtain the content of different chemical elements or compounds>Such as atomic absorption spectroscopy, mass spectrometry, etc.; different chemical elements or compounds are given different weights according to the source of the dredging mud and the characteristics of the environment>Such as according to the extent of its effect on the physical, chemical, biological properties of the dredging mud, etc. Calculating influence coefficient of dredging mud source and environment on chemical components according to formula>. The formula is applied in actual work, and the chemical components of the dredging mud can be regulated by measuring and controlling the influence coefficient of the source and the environment of the dredging mud on the chemical components of the dredging mud, so that the dredging mud is suitable for different recycling modes. For example, if the dredging mud is to be used for ecological restoration, chemical components of origin and environment benefit to it, such as organic matter, nutrient salts, etc., may be selected; if the dredging mud is to be used in other ways, chemical components can be chosen for which the source and environment are not harmful, such as inert substances, stabilizing substances, etc.

The chemical composition of the dredged mud is also related to its structure and morphology, which can result in different chemical compositions, given by:

in the middle of，Is the influence coefficient of the structure and the morphology of the dredging mud on the chemical components; />Is->The content of the seed chemical element or compound; />Is->A score for the structure or morphology of the seed chemical element or compound; />Is the number of kinds of chemical elements or compounds. The parameter acquisition process comprises the following steps: carrying out structural and morphological analysis on a certain amount of dredging mud to obtain structural or morphological information of different chemical elements or compounds, such as a crystal structure, a crystal face orientation, a crystal grain size and the like; according to the influence degree of the structure or the morphology of different chemical elements or compounds on the physical property, the chemical property and the biological property of the dredging mud, different scores are given to different structures or morphologies>Such as according to the extent of its effect on the stability, reactivity, bioavailability, etc. of the dredging mud. Calculating influence coefficient of structure and morphology of dredging mud on chemical components according to formula>. The formula is applied in actual work, and the chemical components of the dredging mud can be regulated by measuring and controlling the influence coefficient of the structure and the form of the dredging mud on the chemical components of the dredging mud, so that the dredging mud is suitable for different recycling modes. For example, if the dredging mud is to be used in building material preparation, the chemical composition for which the structure and morphology are advantageous can be selected Such as stable crystal structure, ordered crystal plane orientation, proper grain size, etc.; if the dredged sediment is to be used in other ways, chemical components can be chosen for which the structure and morphology are not detrimental, such as amorphous structures, irregular morphology, fine particles, etc.

The chemical composition of dredged mud is also related to its time and space, which can lead to different chemical compositions, given by the formula:

in the method, in the process of the invention,the influence coefficient of the time and the space of the dredging mud on the chemical components of the dredging mud; />Is->The content of the seed chemical element or compound; />Is->A score for time of the species chemical element or compound; />Is->A score for the space of the seed chemical element or compound; />Is the number of kinds of chemical elements or compounds. The parameter acquisition process comprises the following steps: carrying out time and space analysis on a certain amount of dredging mud to obtain time and space distribution of different chemical elements or compounds of the dredging mud, such as sampling time, sampling place, sampling depth and the like; according to the differencesThe degree of influence of the temporal and spatial distribution of chemical elements or compounds on the physical, chemical and biological properties of the dredging mud gives different scores to different times and spaces >And->Such as the degree of influence on the dredging mud according to the change trend, the regional difference, the vertical layering and the like of the dredging mud; calculating the influence coefficient of the time and space of dredging mud on its chemical composition according to the formula +.>. The formula is applied in actual work, and the chemical components of the dredging mud can be regulated by measuring and controlling the influence coefficient of the time and the space of the dredging mud on the chemical components of the dredging mud, so that the dredging mud is suitable for different recycling modes. For example, if dredging mud is to be used for ecological restoration, chemical compositions for which time and space are advantageous, such as seasonal variations, regional differences, water depth stratification, etc., may be selected; if the dredged mud is to be used in other ways, chemical compositions that are harmless to it in time and space can be chosen, such as constant, evenly distributed, free of stratification, etc.

The pollution evaluation of dredging mud is related to the type and quantity of pollutants, the more and more harmful the pollutants, the higher the pollution evaluation, the formula is:

in the method, in the process of the invention,evaluating an index for dredging mud pollution; />Is->The content of seed contaminants; />Is->Toxicity coefficient of the seed contaminant; />Is the number of types of contaminants. The parameter acquisition process comprises the following steps: subjecting a quantity of dredging mud to pollutant analysis to obtain different pollutant contents >Such as heavy metals, organics, radioactive materials, etc.; according to the toxicity coefficient of different pollutants +.>Constants related to the degree of damage of the dye, such as the degree of influence on human health, ecological environment, resource utilization, etc. according to the degree of influence; calculating pollution evaluation index of dredging mud according to formula>. The formula is applied to actual work, and the pollution degree of the dredging mud can be adjusted by measuring and controlling the pollution evaluation index of the dredging mud, so that the formula is suitable for different recycling modes. For example, if the dredged mud is to be used for ecological restoration, the dredged mud with low pollution evaluation index can be selected, so that the pollution to the water body is reduced; if the dredging mud is to be used in other ways, a dredging mud with a high pollution evaluation index can be selected, increasing its resource value.

The pollution assessment of dredged mud is also related to the migration and conversion of its pollutants, the easier the pollutants migrate and convert, the higher the pollution assessment, the formula:

in the method, in the process of the invention,migration and conversion coefficients for dredging mud contaminants; />Is->The content of seed contaminants; />Is the firstMigration rate of seed contaminants; />Is->Conversion rate of the seed contaminant; />Is the number of types of contaminants. The parameter acquisition process comprises the following steps: migration and transformation analysis of a certain amount of dredging mud to obtain migration rate of different pollutants >And transformation Rate->Such as dissolution, adsorption, sedimentation, redox, etc.; according to the influence degree of migration and transformation of different pollutants on physical property, chemical property and biological property of dredging mud, different migration and transformation are endowed with different rates>And->Such as according to the degree of its effect on the fluidity, stability, bioavailability, etc. of the dredging mud; calculation of migration and conversion of contaminants of dredged mud according to formulaCoefficient->. The formula is applied to actual work, and the pollution degree of the dredging mud can be regulated by measuring and controlling the migration and conversion coefficient of pollutants of the dredging mud, so that the formula is suitable for different recycling modes. For example, if the dredged mud is to be used for ecological restoration, a dredged mud with low migration and conversion coefficients of pollutants can be selected, so that the pollution of the water body by the dredged mud is reduced; if the dredging mud is to be used in other ways, a dredging mud with a high migration and conversion coefficient of the contaminants can be selected, increasing its resource value.

Biological analysis of a dredged sediment sample, comprising:

the biological effect of dredging mud is related to the kind and quantity of microorganisms, the more and more active the microorganisms, the greater the biological effect, the formula is:

In the method, in the process of the invention,the microbial count of the dredging mud; />Is the initial microbial population; />Is the microorganism growth rate; />Is time. The parameter acquisition process comprises the following steps: performing microorganism analysis on a certain amount of dredging mud to obtain different microorganism types and amounts, such as bacteria, fungi, algae, etc.; according to the growth conditions and growth curves of different microorganisms, the initial quantity +.>And growth rate->Such as according to its requirements and reactions for temperature, light, oxygen, etc.; calculating the number of microorganisms of the dredging mud according to the formula +.>. The formula is applied in actual work, and the biological influence of the dredging mud can be regulated by measuring and controlling the number of microorganisms in the dredging mud, so that the formula is suitable for different recycling modes. For example, if the dredging mud is to be used for ecological restoration, a dredging mud with a high number of microorganisms can be selected to increase its biological activity on the body of water; if the dredging mud is to be used in other ways, a dredging mud with a low number of microorganisms can be selected, reducing its biological consumption of the body of water.

The biological effects of dredging mud are also related to the metabolic activity of its microorganisms, the more the microorganisms consume or produce organic or inorganic matter, the greater the biological effects are expressed as:

In the method, in the process of the invention,microbial metabolic activity for dredging mud; />A change in concentration of a substrate or product that is consumed or produced by the microorganism over a period of time; />Is a time interval. The parameter acquisition process comprises the following steps: subjecting a quantity of dredging mud to metabolic activity analysis to obtain the concentration variation of metabolites or consumables of different microorganisms>Such as organic carbon, nitrogen, phosphorus, etc.; according to the measuring method of different metabolites or consumers, the measuring time interval is calculated>Such as according to its measuring instrument, measuring step, measuring accuracy, etc.; calculating the metabolic activity of the microorganism of the dredging mud according to the formula +.>. The formula is applied in actual work, and the biological influence of the dredging mud can be regulated by measuring and controlling the metabolic activity of microorganisms of the dredging mud, so that the formula is suitable for different recycling modes. For example, if dredging mud is to be used for ecological restoration, dredging mud with high metabolic activity of microorganisms can be selected to increase the biological circulation of the water body; if the dredging mud is to be used in other ways, the dredging mud with low metabolic activity of microorganisms can be selected, so that the biological interference of the dredging mud on the water body is reduced.

In this embodiment, the characteristic analysis of the dredging mud includes physical analysis, chemical analysis and biological analysis; physical analysis includes fluidity and stability, fluidity includes water content, average particle diameter and temperature influence coefficient; stability includes inter-particle van der waals forces, particle porosity, and particle young's modulus; chemical analysis includes chemical composition and pollution assessment, the chemical composition including source and environmental impact coefficients, structural and morphological impact coefficients and temporal and spatial impact coefficients; pollution assessment includes pollution assessment index, migration and conversion coefficient; biological assays include biological effects, including microbial numbers and microbial metabolic activity; and according to the actual situation, specifically analyzing what kind of characteristics are needed as a resource utilization mode of the dredging mud for subsequent analysis.

In fig. 2-3, conv2D and Conv1D each represent a standard convolution layer, with convolution kernel sizes 3, 7, and 1; strides represents the step size, and takes on the value of 1 or 2; the normalized Activation layer comprises a batch normalization layer (Batch Normalization) and an Activation function layer (Activation (Relu)), wherein the normalized Activation layer selects a Relu Activation function, a single batch normalization layer (Batch Normalization), and a single Activation function layer (Activation #)))，/>The value is Relu; separablecon v1D represents a depth separable one-dimensional convolution layer with a convolution kernel size of 3; globalaeragepooling 2D and globalagepooling 1D each represent a global average pooling layer; maxpooling2D represents the maximum pooling layer; reshape stands for tensor remodelling layer; dense stands for full connectivity layer; add (/ -)>,/>) Representation->，/>Performing element-by-element addition; concat (+)>,) Representation->，/>Performing tensor stitching; s->Representing the resulting feature map->The value range is [1,30 ]]，/>Is an integer.

The step S3 specifically comprises the following steps:

sequentially inputting dredging mud sample images into a first standard convolution layer, a first normalized activation layer and a first maximum pooling layer to carry out rolling and pooling operation to obtain a feature map S3, wherein the feature map S3 specifically comprises:

Inputting the dredging mud sample image (256,256,3) or (256,256,1) into a first standard convolution layer for convolution operation to obtain a feature map S1, wherein the number of convolution kernels of the first standard convolution layer is 64, the size of the convolution kernels is 7 multiplied by 7, and the step length is 2; the feature map S1 is 128×128 of 64 channels; inputting the feature map S1 into a first normalized activation layer to perform batch normalization and activation operation to obtain a feature map S2; the feature map S2 is 128×128 of 64 channels; inputting the feature map S2 into a first maximum pooling layer to perform maximum pooling operation to obtain a feature map S3; the size of the pooling window of the first maximum pooling layer is 3 multiplied by 3, and the step length is 2; the feature map S3 is 64×64 of 64 channels.

The feature map S3 is sequentially input into a first structured dynamic convolution module, a second normalized activation layer, a second structured dynamic convolution module and a third normalized activation layer to perform dynamic convolution operation, so as to obtain a feature map S7, which specifically includes:

the feature map S3 is sequentially input to a first structured dynamic convolution module, where the first structured dynamic convolution module uses an output channel number of 64, a convolution kernel size of 7, and a dynamic scaling factor of 8, and specifically includes:

calculating the average value of the feature map S3 in the height and width dimensions to obtain dynamic input data, including:

First, the input tensor (e.g., an image, feature map) is averaged over the height and width dimensions to obtain a simplified data representation. This representation captures the overall characteristics of the input data and is referred to as "dynamic input data".

Generating weights of dynamic input data by using a full connection layer to obtain dynamic convolution kernel weights, wherein the method specifically comprises the following steps:

this dynamic input data is fed into a fully connected layer. This fully connected layer generates weights based on dynamic input data, which will be used as convolution kernel weights in the convolution operation. These weights are called "dynamic convolution kernel weights" because they are generated from each input change.

Obtaining the height and width of the convolution kernel size according to the convolution kernel size, wherein the input channel number=the final dimension/dynamic scale factor of the input tensor (reducing the input channel number scale factor); the output dimension of the fully-connected layer is the height of the convolution kernel size x the width of the convolution kernel size x the number of input channels x the number of output channels.

Remolding the dynamic convolution kernel weight to obtain a dynamic convolution kernel, which specifically comprises the following steps:

the generated dynamic convolution kernel weights are then reshaped into a shape suitable for the convolution operation. This typically involves adjusting the weights to a four-dimensional shape to fit the standard dimension of the convolution kernel: height, width, number of input channels, and number of output channels.

Performing convolution operation on the input tensor by using a dynamic convolution kernel to obtain an output tensor, wherein the convolution operation specifically comprises the following steps:

a convolution operation is performed using these remolded dynamic convolution check input tensors. This step generates the final output tensor (feature map S4), i.e., the convolved data.

Inputting the feature map S4 into a second normalized activation layer to obtain a feature map S5; the feature map S5 is 64×64 of 64 channels.

And sequentially inputting the feature map S5 into a second structured dynamic convolution module and a third normalized activation layer to perform dynamic convolution operation to obtain a feature map S7, wherein the principle of the second structured dynamic convolution module is the same as that of the first structured dynamic convolution module, the parameters are consistent, and the feature map S7 is 64 multiplied by 64 of 64 channels.

The feature map S7 is sequentially input to a ternary attention module and a multiscale recursive attention module for performing attention operation, so as to obtain a feature map S9, which specifically includes:

inputting the feature map S7 to a ternary attention module to obtain a feature map S8, which specifically includes:

using a global average pooling layer and a global maximum pooling layer to pool input features (feature map S7), and respectively obtaining global average pooling features and global maximum pooling features through learning channel attention of a full connection layer; the full connection layer I is used for learning channel attention from the global average pooled characteristics, and the output dimension of the full connection layer I is the number of input characteristic channels divided by the compression ratio, and the ratio is 8; this compression ratio is used to calculate the dimension of the compression feature when the channel is attentive; adding the global average pooling feature and the global maximum pooling feature, and recovering the dimension through the full connection layer II to obtain the channel attention weight; the full connection layer II is used for recovering the dimension of the channel attention to the number of input characteristic channels; the full connection layer I and the full connection layer II use a ReLU activation function and do not use bias items; and adjusting the shape of the channel attention weight to be the same as the shape of the input feature map to obtain the channel attention.

Calculating spatial attention using a two-dimensional convolution layer; the two-dimensional convolution layer is used for realizing a spatial attention mechanism, the convolution kernel size of the two-dimensional convolution layer is 7, the activation function is Sigmoid, and no offset term is used.

Multiplying the calculated spatial attention and channel attention by the input features, and outputting a feature map S8; the feature map S8 is 64×64 of 64 channels; enhancing the interest of the network for important features.

Channel attention, focusing on features on different channels through global average pooling and global maximum pooling, and then learning which channels are important through fully connected layers; spatial attention, using convolutional layers to focus on different locations in the input feature map, learning which spatial regions are important; the combined attention, the channel attention and the spatial attention are combined and applied to the input features by element-wise multiplication, thereby focusing on important channel and spatial regions simultaneously.

Inputting the feature map S8 into a multi-scale recursive attention module to obtain a feature map S9, which specifically includes:

initializing an empty list (attention map list) for storing attention maps on different scales;

setting a target size, and acquiring the height and the width of a target feature map from the target size; traversing the scale list, creating a corresponding number of ternary attention module examples according to the number of scales in the scale list, and operating the input feature map of each scale, wherein the method specifically comprises the following steps:

Carrying out average pooling on the input feature map by using an average pooling layer, wherein the pooling size is determined by the current scale; repeatedly applying a ternary attention module to the pooled feature map, and setting the recursion times according to actual conditions; adjusting the processed feature map to a target size; adding the resized feature graphs to an attention profile list; combining attention diagrams on all scales to obtain a final output characteristic diagram.

In this embodiment, the instances in the list of ternary attention module instances are created from the number of elements in the scale list. When referring to "first", "second", "third" ternary attention module instances, these ternary attention module instances are ordered in the list. If the scale list= [1,2,4], three instances will be created, corresponding to the three scales, respectively, specifically: the first ternary attention module instance corresponds to the first element in the scale list (scale 1), this instance being used to process the input feature map of the original scale (i.e., without a size transformation); a second ternary attention module instance corresponds to the second element in the scale list (scale 2), this instance being used to process the input feature map scaled down to half the original size; a third ternary attention module instance corresponds to the third element in the scale list (scale 4), this instance being used to process input feature maps scaled down to one-fourth of the original size.

Assuming an input feature map of size (128) pixels, the scale list is set to [1,2,4], and the number of recursions is set to 2. This means that the attention mechanism will be applied on three different scales and repeated twice on each scale; for scale 1, without pooling, applying the ternary attention module directly on the original (128 ) feature map will result in an attention map of (128 ) since the number of recursions is 2, and therefore the same ternary attention module will be applied twice in succession; for scale 2, the input feature map is first (2, 2) averaged pooled, which reduces the feature map size by half to (64, 64), and then a second ternary attention module is applied twice in succession to this (64, 64) feature map, resulting in an attention map of (64, 64); for scale 4, the input feature map is first (4, 4) averaged pooled, which reduces the feature map size by a factor of four to (32, 32), and then a third ternary attention module is applied twice in succession on this (32, 32) feature map, resulting in an attention map of (32, 32).

After processing at each scale, attention at all scales is required to be drawn to the same target size, here assuming that the target size is (64, 64); thus, the attention of (128 ) is intended to be downsampled to (64, 64), and the attention of (32, 32) is intended to be upsampled to (64, 64); finally, the attention patterns of the three (64, 64) are combined to obtain a final output characteristic map S9.

The multi-scale recursive attention module can capture characteristic information of different scales by repeatedly applying attention mechanisms on different scales, and is beneficial to improving the understanding capability of the model on input data.

The feature map S9 is sequentially input to a second standard convolution layer, a fourth normalized activation layer, a third standard convolution layer, a first normalization layer, a fourth standard convolution layer, a second normalization layer and an element addition layer to perform convolution and element addition operations, so as to obtain a feature map S16, which specifically includes:

inputting the feature map S9 into a second standard convolution layer to carry out convolution operation to obtain a feature map S10, wherein the number of convolution kernels of the second standard convolution layer is 128, the size of the convolution kernels is 3 multiplied by 3, and the step length is 2; the feature map S10 is 32×32 of 128 channels; inputting the feature map S10 into a fourth normalized activation layer to perform batch normalization and activation operation to obtain a feature map S11; the feature map S11 is 32×32 of 128 channels; inputting the feature map S11 into a third standard convolution layer for convolution operation to obtain a feature map S12, wherein the number of convolution kernels of the third standard convolution layer is 128, the size of the convolution kernels is 3 multiplied by 3, and the step length is 1; the feature map S12 is 32×32 of 128 channels; inputting the feature map S12 into a first normalization layer for batch normalization operation to obtain a feature map S13; the feature map S13 is 32×32 of 128 channels; inputting the feature map S9 into a fourth standard convolution layer for convolution operation to obtain a feature map S14, wherein the number of convolution kernels of the fourth standard convolution layer is 128, the size of the convolution kernels is 1 multiplied by 1, and the step length is 2; the feature map S14 is 32×32 of 128 channels; inputting the feature map S14 into a second batch normalization layer to perform batch normalization operation to obtain a feature map S15; feature map S15 is 32×32 of 128 channels; the feature map S13 and the feature map S15 are input to an element addition layer to perform element addition operation, and a feature map S16 is obtained.

The feature map S16 is sequentially input to a first activation function layer, a second maximum pooling layer, a fifth standard convolution layer, a third maximum pooling layer and a first global average pooling layer to perform feature extraction operation, so as to obtain an image feature S21, which specifically includes:

inputting the feature map S16 into a first activation function layer to perform activation function operation to obtain a feature map S17; feature map S17 is 32×32 of 128 channels; inputting the feature map S17 into a second maximum pooling layer to perform maximum pooling operation to obtain a feature map S18; the size of the pooling window of the second maximum pooling layer is 3 multiplied by 3, and the step length is 2; the feature map S18 is 16×16 of 128 channels; inputting the feature map S18 into a fifth standard convolution layer for convolution operation to obtain a feature map S19, wherein the number of convolution kernels of the fifth standard convolution layer is 256, the size of the convolution kernels is 3 multiplied by 3, and the step length is 2; the feature map S19 is 8×8 of 256 channels; inputting the feature map S19 into a third maximum pooling layer for maximum pooling operation to obtain a feature map S20; the size of a pooling window of the third maximum pooling layer is 3 multiplied by 3, and the step length is 2; the feature map S20 is 4×4 of 256 channels; the feature map S20 is input to the first global averaging pooling layer, and the output image features S21, S21 are (256).

In this embodiment, the image feature extraction network includes a first standard convolution layer, a first normalized activation layer, a first maximum pooling layer, a first structured dynamic convolution module, a second normalized activation layer, a second structured dynamic convolution module, a third normalized activation layer, a ternary attention module, a multi-scale recursive attention module, a second standard convolution layer, a fourth normalized activation layer, a third standard convolution layer, a first batch normalization layer, a fourth standard convolution layer, a second batch normalization layer, an element addition layer, a first activation function layer, a second maximum pooling layer, a fifth standard convolution layer, a third maximum pooling layer, and a first global average pooling layer.

preprocessing physical characteristics, chemical characteristics and biological characteristics, and inputting the processed result into a tensor remolding layer for tensor remolding to obtain a characteristic diagram S22; the feature map S22 is (13, 1); inputting the feature map S22 into a depth separable convolution layer to perform depth separable convolution operation to obtain a feature map S23, wherein the number of convolution kernels of the depth separable convolution layer is 64, the size of the convolution kernels is 3, and the step length is 1; the feature map S23 is (13,64); inputting the feature map S23 into a third batch normalization layer to perform batch normalization operation to obtain a feature map S24; the feature map S24 is (13,64); inputting the feature map S24 into a sixth standard convolution layer for convolution operation to obtain a feature map S25, wherein the number of convolution kernels of the sixth standard convolution layer is 128, the size of the convolution kernels is 3, and the step length is 1; the feature map S25 is (13,128); inputting the feature map S25 into a fourth normalization layer for batch normalization operation to obtain a feature map S26; the feature map S26 is (13,128); inputting the feature map S26 into a second global average pooling layer to obtain a feature map S27; s27 is (128), which comprises physical characteristics, chemical characteristics and biological characteristic information; inputting the feature map S21 and the feature map S27 into a tensor splicing layer to perform tensor splicing operation to obtain a feature map S28; the feature map S28 is (384,); the feature map S28 is sequentially input to the full-connection activation layer and the full-connection classification layer, and the resource classification is performed, so that the resource utilization mode is output.

In this embodiment, the resource utilization mode includes building material equipment, sea-filling land-making, ecological restoration and others; the building material equipment mainly utilizes clay, quartz and the like in the dredging mud to prepare substances with ceramic properties, such as bricks, ceramsites, tiles, ceramics and other building materials by sintering or hot pressing, and the like; sea sand, dredged soil and the like in the dredged mud are mainly utilized for sea filling and land making, and are desalted or solidified by a physical or chemical method and used for engineering of sea filling and land making, such as road building, dike building, low-lying area backfilling and the like, so that the space value of the dredged mud can be effectively utilized, the land resource is increased, and the land utilization rate is improved; the ecological restoration mainly utilizes organic matters, nutritive salts and the like in the dredging mud, prepares substances with fertilizer effect through composting or fermentation and the like, and is used for ecological restoration projects such as ecological bank slope construction, submarine pit restoration, mine restoration and the like, the biological properties of the dredging mud can be effectively utilized through the mode, the contribution to the fields of agricultural production, soil improvement, ecological restoration and the like is increased, and the ecological stability and the ecological sustainability are improved; other methods include utilizing biological resources in dredged mud, such as fat, sugar, etc., to convert into combustible energy sources, such as biogas, ethanol, grease, etc., by biochemical or thermochemical methods; the biological resources in the dredging mud, such as proteins, vitamins and the like, are utilized to prepare products with specific functions or purposes, such as enzymes, antibodies and medicines by means of bioengineering or biotechnology; biological resources such as algae and microorganisms in dredging mud are utilized, and substances with feeding value such as algae powder and microorganism powder are prepared by drying or crushing. The type of the resource utilization mode and the specific type of the resource are selected according to the actual situation.

In this embodiment, the resource network includes a tensor remodelling layer, a depth separable convolution layer, a third normalization layer, a sixth standard convolution layer, a fourth normalization layer, a second global average pooling layer, a tensor stitching layer, a full connection activation layer, and a full connection classification layer.

The step S5 specifically comprises the following steps:

the planning scheme should consider the physical, chemical, biological and image characteristics of the dredging mud, and the benefits and risks of the resource utilization mode, select the most suitable treatment and resource utilization technology, formulate detailed operation flow and management measures, and evaluate the economy, feasibility and sustainability of the project.

In the process of engineering projects of sea reclamation, the following should be considered:

ensuring that the product meets relevant environmental standards and safety requirements; when the land is made by filling the sea, reasonable engineering design and construction method are adopted to ensure the quality and stability of the land, so as to avoid the problems of sedimentation, landslide and the like; after the reclamation of land, the environmental impact and social benefits of the reclamation of land should be continuously monitored and evaluated to ensure sustainable development and balance of benefits of reclamation of land.

In the engineering project of ecological restoration, the following should be considered:

ensuring that the plant material meets relevant ecological standards and repair requirements; when ecological restoration is carried out, a reasonable restoration scheme and a reasonable technical method are adopted to ensure the effect and quality of ecological restoration and enhance the functions and services of an ecological system; after ecological restoration, the ecological benefit and the social benefit of the ecological restoration should be continuously monitored and evaluated to ensure the long-acting and adaptability of the ecological restoration.

In the engineering project of building material preparation, the following should be considered:

ensuring that the building materials meet the relevant building material standards and quality requirements; when building material preparation is carried out, reasonable preparation technology and technical method are adopted to ensure the efficiency and quality of building material preparation and improve the performance and added value of the building material; after the building material preparation, the environmental impact and economic benefit of the building material preparation should be continuously monitored and evaluated to ensure the environmental protection and economical efficiency of the building material preparation.

Example 2

As shown in fig. 2, the invention discloses a dredging mud recycling evaluation system based on deep learning, which comprises:

a data sample acquisition module 10 for acquiring a dredged mud sample.

The basic characteristic extraction module 20 is used for performing physical analysis, chemical analysis and biological analysis on the dredged mud sample to obtain physical characteristics, chemical characteristics and biological characteristics.

The image feature extraction module 30 is configured to input the dredging mud sample image into an image feature extraction network for feature extraction, so as to obtain an image feature.

The resource utilization classification module 40 is configured to input the image feature, the physical feature, the chemical feature, and the biological feature into the resource network for identification, and output a resource utilization mode.

The standard scheme making module 50 is configured to make a plan according to the resource utilization mode, so as to obtain a planning scheme.

As an alternative embodiment, the basic characteristic extraction module 20 of the present invention specifically includes:

the formula of the water content of the dredging mud is as follows:

in the method, in the process of the invention,is the water content of the dredging mud; />Is the mass of water; />Is solid mass.

The formula of the average particle diameter of the dredging mud is as follows:

in the method, in the process of the invention,is the average particle diameter of the dredging mud; />Is->Particle diameter; />Is->The individual particle frequencies; />The number of the particles.

in the method, in the process of the invention,is the temperature influence coefficient of the dredging mud; />Viscosity of the dredged mud; />Is the reference viscosity; / >Is the activation energy;is a gas constant; />Is the temperature of the dredging mud; />Is the reference temperature.

In the method, in the process of the invention,van der Waals forces between dredging mud particles; />Is a hamilton constant; />Is the thickness of the dredging mud; />Is the radius of the particles; />Is the inter-particle distance.

The formula of the dredged mud porosity is as follows:

in the method, in the process of the invention,is the dredged mud porosity; />Is the total volume of the dredging mud; />Is the volume of dredged mud solids.

The Young modulus formula of the dredging mud is as follows:

in the method, in the process of the invention,young's modulus for dredging mud particles; />Stress of the particles; />Is the strain of the particles.

in the method, in the process of the invention,is the dredging mud source and the environmental impact coefficient; />Is->The content of the seed compound; />Is->Weights of seed compounds; />Number of kinds of compounds.

in the method, in the process of the invention,is the structure and form influence coefficient of the dredging mud; />Is->The content of the seed compound; />Is->A structural or morphological score of the seed compound; />The number of kinds of the compounds.

in the method, in the process of the invention,is the dredging mud time and space influence coefficient; />Is->The content of the seed compound; />Is->Time score of seed compound; / >Is->Spatial score of the seed compound; />The number of kinds of the compounds.

The dredging mud pollution evaluation index formula is:

in the method, in the process of the invention,evaluating an index for dredging mud pollution; />Is->Seed pollutant content; />Is->Toxicity coefficient of seed pollutant; />Is the number of contaminant species.

in the method, in the process of the invention,migration and conversion coefficients for the dredged mud; />Is->Seed pollutant content; />Is->Seed contaminant migration rate; />Is->Rate of conversion of the seed contaminant; />Is the number of contaminant species.

The formula of the microorganism number of the dredging mud is as follows:

in the method, in the process of the invention,the microbial count of the dredging mud; />Is the initial microbial population; />Is the microorganism growth rate; />Is time.

in the method, in the process of the invention,microbial metabolic activity for dredging mud; />A change in concentration of a substrate or product that is consumed or produced by the microorganism over a period of time; />Is a time interval. As an alternative embodiment, the image feature extraction module 30 of the present invention specifically includes:

the first standard convolution module is used for inputting the dredging mud sample image into the first standard convolution layer, the first normalized activation layer and the first maximum pooling layer in sequence to carry out convolution and pooling operation, and a characteristic diagram S3 is obtained.

The structured dynamic convolution module is configured to input the feature map S3 to the first structured dynamic convolution module, the second normalized activation layer, the second structured dynamic convolution module, and the third normalized activation layer in order to perform a dynamic convolution operation, so as to obtain a feature map S7.

The attention mechanism module is used for sequentially inputting the feature map S7 into the ternary attention module and the multi-scale recursive attention module to perform attention operation, so as to obtain a feature map S9.

The residual module is used for inputting the feature map S9 into the second standard convolution layer, the fourth normalized activation layer, the third standard convolution layer, the first normalization layer, the fourth standard convolution layer, the second normalization layer and the element addition layer in sequence to carry out convolution and element addition operation, so as to obtain the feature map S16.

The image feature output module is used for sequentially inputting the feature map S16 into the first activation function layer, the second maximum pooling layer, the fifth standard convolution layer, the third maximum pooling layer and the first global average pooling layer to perform feature extraction operation, so as to obtain image features S21.

As an alternative embodiment, the first structured dynamic convolution module of the present invention specifically includes:

and the input sub-module is used for calculating the average value of the input tensor in the height dimension and the width dimension to obtain dynamic input data.

And the dynamic convolution kernel weight generation sub-module is used for generating the weight of the dynamic input data by using the full connection layer to obtain the dynamic convolution kernel weight.

And the dynamic convolution kernel acquisition submodule is used for remolding the dynamic convolution kernel weight to obtain a dynamic convolution kernel.

And the output sub-module is used for carrying out convolution operation on the input tensor by using the dynamic convolution check to obtain an output tensor.

As an alternative embodiment, the multi-scale recursive attention module of the present invention specifically comprises:

and the initialization sub-module is used for initializing the attention map list and setting the target size.

The multi-scale operation sub-module is used for traversing the scale list, creating a corresponding number of ternary attention modules according to the number of scales in the scale list, and operating the input feature map of each scale, and specifically comprises the following steps:

carrying out average pooling on the input feature map by using an average pooling layer; repeatedly applying a ternary attention module to the pooled feature map; adjusting the processed feature map to a target size; adding the resized feature graphs to an attention profile list; combining attention diagrams on all scales to obtain a final output characteristic diagram.

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

1. The dredging mud resource utilization evaluation method based on deep learning is characterized by comprising the following steps of:

step S1: acquiring a dredging mud sample;

step S2: performing physical analysis, chemical analysis and biological analysis on the dredging mud sample to obtain physical characteristics, chemical characteristics and biological characteristics, wherein the physical characteristics, the chemical characteristics and the biological characteristics specifically comprise:

the physical characteristics include moisture content, average particle diameter, temperature influence coefficient, particle van der Waals force, porosity and Young's modulus, the chemical characteristics include source and environmental influence coefficient, structure and morphology influence coefficient, time and space influence coefficient, migration and conversion coefficient and pollution evaluation index, the biological characteristics include microbial number and microbial metabolic activity;

the formula of the water content of the dredging mud is as follows:

；

the formula of the average particle diameter of the dredging mud is as follows:

；

in the method, in the process of the invention,is the average particle diameter of the dredging mud; />Is->Particle diameter; />Is->The individual particle frequencies; />The number of the particles;

；

in the method, in the process of the invention,is the temperature influence coefficient of the dredging mud; />Is dredgingMud viscosity; />Is the reference viscosity; / >Is the activation energy; />Is a gas constant; />Is the temperature of the dredging mud; />Is the reference temperature;

the Van der Waals force formula between dredging mud particles is:

；

the formula of the dredged mud porosity is as follows:

；

the Young modulus formula of the dredging mud is as follows:

；

in the method, in the process of the invention,is the structure and form influence coefficient of the dredging mud; />Is->A structural or morphological score of the seed compound;

；

in the method, in the process of the invention,is the dredging mud time and space influence coefficient; />Is->Time score of seed compound; />Is->Spatial score of the seed compound;

；

in the method, in the process of the invention,migration and conversion coefficients for dredged mud; />Is->Seed pollutant content; />Is->Seed contaminant migration rate; />Is->Rate of conversion of the seed contaminant; />The number of contaminant species;

the dredging mud pollution evaluation index formula is:

；

in the method, in the process of the invention,evaluating an index for dredging mud pollution; />Is->Toxicity coefficient of seed pollutant;

the formula of the microorganism number of the dredging mud is as follows:

；

in the method, in the process of the invention,microbial metabolic activity for dredging mud; />A change in concentration of a substrate or product that is consumed or produced by the microorganism over a period of time; />Is a time interval;

step S3: inputting a dredging mud sample image into an image feature extraction network for feature extraction to obtain image features, wherein the method specifically comprises the following steps of:

the image feature extraction network comprises a first standard convolution layer, a first normalized activation layer, a first maximum pooling layer, a first structured dynamic convolution module, a second normalized activation layer, a second structured dynamic convolution module, a third normalized activation layer, a ternary attention module, a multi-scale recursive attention module, a second standard convolution layer, a fourth normalized activation layer, a third standard convolution layer, a first batch normalization layer, a fourth standard convolution layer, a second batch normalization layer, an element addition layer, a first activation function layer, a second maximum pooling layer, a fifth standard convolution layer, a third maximum pooling layer and a first global average pooling layer;

Sequentially inputting the dredging mud sample image into the first standard convolution layer, the first normalized activation layer and the first maximum pooling layer to perform rolling and pooling operation to obtain a feature map S3;

inputting the feature map S3 into the first structured dynamic convolution module, the second normalized activation layer, the second structured dynamic convolution module and the third normalized activation layer in sequence to perform dynamic convolution operation to obtain a feature map S7;

inputting the feature map S7 to the ternary attention module and the multi-scale recursive attention module in sequence to perform attention operation to obtain a feature map S9;

inputting the feature map S9 into the second standard convolution layer, the fourth normalized activation layer, the third standard convolution layer, the first normalization layer, the fourth standard convolution layer, the second normalization layer and the element addition layer in sequence to perform convolution and element addition operation to obtain a feature map S16;

inputting the feature map S16 into the first activation function layer, the second maximum pooling layer, the fifth standard convolution layer, the third maximum pooling layer and the first global average pooling layer in sequence to perform feature extraction operation to obtain image features S21;

Step S4: inputting the image features, the physical features, the chemical features and the biological features into a resource network for identification, and outputting a resource utilization mode, wherein the method specifically comprises the following steps of:

the resource network comprises a tensor remodelling layer, a depth separable convolution layer, a third normalization layer, a sixth standard convolution layer, a fourth normalization layer, a second global average pooling layer, a tensor splicing layer, a full-connection activation layer and a full-connection classification layer;

preprocessing the physical characteristics, the chemical characteristics and the biological characteristics, and inputting the processed results into the tensor remolding layer for tensor remolding to obtain a characteristic diagram S22; inputting the characteristic map S22 into the depth separable convolution layer to perform depth separable convolution operation to obtain a characteristic map S23; inputting the feature map S23 to the third normalization layer for batch normalization operation to obtain a feature map S24; inputting the feature map S24 to the sixth standard convolution layer for convolution operation to obtain a feature map S25; inputting the feature map S25 to the fourth normalization layer for batch normalization operation to obtain a feature map S26; inputting the feature map S26 to the second global average pooling layer to obtain a feature map S27; inputting the image features S21 and the feature map S27 into the tensor splicing layer for tensor splicing operation to obtain a feature map S28; the feature map S28 is sequentially input into the full-connection activation layer and the full-connection classification layer to carry out resource classification and output a resource utilization mode;

2. The deep learning-based dredging mud recycling evaluation method according to claim 1, wherein the first structured dynamic convolution module specifically comprises:

3. Deep learning-based dredging mud recycling evaluation method according to claim 1, wherein the multi-scale recursive attention module specifically comprises:

initializing an attention map list and setting a target size;

4. Deep learning-based dredging mud recycling evaluation system is characterized in that the system comprises:

the data sample acquisition module is used for acquiring a dredging mud sample;

the basic characteristic extraction module is used for carrying out physical analysis, chemical analysis and biological analysis on the dredging mud sample to obtain physical characteristics, chemical characteristics and biological characteristics, and specifically comprises the following steps:

the formula of the water content of the dredging mud is as follows:

；

the formula of the average particle diameter of the dredging mud is as follows:

；

in the method, in the process of the invention,is the average particle diameter of the dredging mud; />Is->Particle diameter; />Is->The individual particle frequencies;/>the number of the particles;

；

in the method, in the process of the invention,is the temperature influence coefficient of the dredging mud; / >Viscosity of the dredged mud; />Is the reference viscosity; />Is the activation energy; />Is a gas constant; />Is the temperature of the dredging mud; />Is the reference temperature;

the Van der Waals force formula between dredging mud particles is:

；

the formula of the dredged mud porosity is as follows:

；

the Young modulus formula of the dredging mud is as follows:

；

in the method, in the process of the invention,is the dredging mud time and space influence coefficient; />Is->Time score of seed compound; / >Is->Spatial score of the seed compound;

；

the dredging mud pollution evaluation index formula is:

；

the formula of the microorganism number of the dredging mud is as follows:

；

the image feature extraction module is used for inputting the dredging mud sample image into the image feature extraction network to perform feature extraction to obtain image features, and specifically comprises the following steps:

The first standard convolution module is used for inputting dredging mud sample images into the first standard convolution layer, the first normalized activation layer and the first maximum pooling layer in sequence to carry out convolution and pooling operations to obtain a feature map S3;

the residual error module is used for inputting the feature map S9 into the second standard convolution layer, the fourth normalized activation layer, the third standard convolution layer, the first normalization layer, the fourth standard convolution layer, the second normalization layer and the element addition layer in sequence to carry out convolution and element addition operation to obtain a feature map S16;

the image feature output module is used for sequentially inputting the feature map S16 into the first activation function layer, the second maximum pooling layer, the fifth standard convolution layer, the third maximum pooling layer and the first global average pooling layer to perform feature extraction operation to obtain image features S21;

The resource utilization classification module is used for inputting the image features, the physical features, the chemical features and the biological features into a resource network for identification, and outputting a resource utilization mode, and specifically comprises the following steps:

5. The deep learning-based dredging mud recycling evaluation system according to claim 4, wherein the first structured dynamic convolution module specifically comprises:

6. Deep learning-based dredging mud recycling evaluation system according to claim 4, characterized in that the multi-scale recursive attention module comprises in particular: