CN114511791A - Regional water resource classification evaluation method based on improved deep residual error network - Google Patents
Regional water resource classification evaluation method based on improved deep residual error network Download PDFInfo
- Publication number
- CN114511791A CN114511791A CN202210413863.6A CN202210413863A CN114511791A CN 114511791 A CN114511791 A CN 114511791A CN 202210413863 A CN202210413863 A CN 202210413863A CN 114511791 A CN114511791 A CN 114511791A
- Authority
- CN
- China
- Prior art keywords
- image
- remote sensing
- water resource
- residual error
- regional water
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 101
- 238000011156 evaluation Methods 0.000 title claims abstract description 17
- 230000004927 fusion Effects 0.000 claims abstract description 15
- 150000001875 compounds Chemical class 0.000 claims abstract description 5
- 238000000034 method Methods 0.000 claims description 25
- 238000003709 image segmentation Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 8
- 230000003595 spectral effect Effects 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 7
- 238000004458 analytical method Methods 0.000 abstract description 5
- 230000000007 visual effect Effects 0.000 abstract description 4
- 238000005070 sampling Methods 0.000 abstract description 3
- 238000004422 calculation algorithm Methods 0.000 description 24
- 238000007781 pre-processing Methods 0.000 description 22
- 239000010410 layer Substances 0.000 description 17
- 230000006870 function Effects 0.000 description 15
- 238000013507 mapping Methods 0.000 description 14
- 238000007635 classification algorithm Methods 0.000 description 12
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 10
- 230000001965 increasing effect Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 9
- 238000011161 development Methods 0.000 description 6
- 230000018109 developmental process Effects 0.000 description 6
- 239000011229 interlayer Substances 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 230000015556 catabolic process Effects 0.000 description 3
- 238000006731 degradation reaction Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 239000002689 soil Substances 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 101000912874 Schizosaccharomyces pombe (strain 972 / ATCC 24843) Iron-sensing transcriptional repressor Proteins 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000007500 overflow downdraw method Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 241000251468 Actinopterygii Species 0.000 description 1
- 241000237536 Mytilus edulis Species 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000013505 freshwater Substances 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 235000020638 mussel Nutrition 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 238000003911 water pollution Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- 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
- Y02A20/00—Water conservation; Efficient water supply; Efficient water use
- Y02A20/152—Water filtration
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Educational Administration (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- General Engineering & Computer Science (AREA)
- Primary Health Care (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Image Processing (AREA)
Abstract
The invention relates to the technical field of depth residual error networks, and discloses a regional water resource classification evaluation method based on an improved depth residual error network. The regional water resource classification evaluation method based on the improved depth residual error network is characterized in that image enhancement is to improve image information and quality, so that characteristics of the regional water resource classification evaluation method are more obvious under human eye observation, the regional water resource identification effect is enhanced, digital interference degree is ignored in visual analysis, an important role is played in all aspects of remote sensing images, image fusion is to carry out replay sampling on a multispectral image with low spatial resolution and a single-waveband image with high spatial resolution to generate a new multispectral image with high resolution, so that the generated image has high spatial resolution and multispectral characteristics, a compound mode is mainly adopted, information provided by remote sensing image data sources of different sensors is integrated, and high-quality image information is obtained.
Description
Technical Field
The invention relates to the technical field of deep residual error networks, in particular to a regional water resource classification evaluation method based on an improved deep residual error network.
Background
Water is a life source, is a key for production and is a basis of ecology, one of strategic measures for accelerating the transformation of economic development mode in China is strict water resource management, three lines of water resource development and utilization control, water use efficiency control and water function limitation are established and definite, and a water resource management responsibility and evaluation system is established, which is the most strict water resource management system in China historically, China is a country suffering from severe drought and water shortage, and the total amount of fresh water resources is 2.8 trillion m3It accounts for 6% of the total amount of the water resources per capita in the world, but the water resources per capita are very low and account for only one fourth of the average level in the world, and are 2200 cubic meters, which is one of 13 poor countries in the world.
With the rapid development of regional socioeconomic, the over expansion of population and the serious aggravation of water pollution, the contradiction between supply and demand of regional water resources is increasingly obvious, the sustainable utilization of regional water resources becomes the core problem of the sustainable development of socioeconomic, China has a plurality of large rivers, the river basin of which often covers a plurality of different geographical or administrative regions, and the great difference exists in the aspects of terrain, climate conditions, soil conditions, vegetation conditions, population density, socioeconomic development level and the like, meanwhile, the water resource conditions are continuously changed along with the change of global climate and the development of socioeconomic, the water resource storage and supply in China are extremely limited, the regional difference is large, the east-west distribution is unbalanced, the north is especially serious water shortage in the northwest and the north, in order to explore the conditions of water resources in different regions, the conditions of water resources in different regions often need to be evaluated, Measuring and classifying, but the existing water resource measuring and calculating method has the defects of long image preprocessing time, high classification error rate and long classification time.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a regional water resource classification evaluation method based on an improved deep residual error network, which has the advantages of low regional water resource classification error rate and the like, and solves the problems of long image preprocessing time, high classification error rate, long classification time and the like.
In order to achieve the purpose of low error rate of regional water resource classification, the invention provides the following technical scheme: a regional water resource classification evaluation method based on an improved depth residual error network comprises four steps of image enhancement, image fusion, image marking and image segmentation;
the specific steps of the image enhancement and the Gaussian low-pass filtering are as follows:
(1) selecting a remote sensing image with obvious water and land resolution, and reading all four wave bands of the image in a gray scale image mode;
(2) an independent Gaussian filter blurs the edge of the remote sensing image to reduce the contrast;
(3) directly moving a Gaussian kernel to a pixel to be processed in a remote sensing image;
(4) multiplying the pixel values of the corresponding positions of the remote sensing images by the weight of the Gaussian kernel respectively;
(5) and adding the result obtained in the last step into a pixel value remote sensing image of the area image.
Furthermore, the image fusion mainly adopts a compound mode, integrates information provided by remote sensing image data sources of different sensors, obtains high-quality image information, eliminates information difference among a plurality of sensors, reduces the fuzzy degree of the remote sensing image, and improves the definition of the image.
Further, the image markers have primitive features including spectral average, area, squareness, and aspect ratio.
Further, the image marking is divided into the following steps:
(1) by utilizing the merging function of OpenCV, the first three wave bands of the original remote sensing image are selected as three channels of BGR, and a new color image presented in a jpg format is synthesized;
(2) marking the water part in the color image in a polygonal form by using a Labelme image marking tool, and storing the vertexes of the marked polygon during marking;
(3) creating an image with the same size as the original image, defaulting to black, reading the vertex of each polygon, and filling the RGB of pixel points in the polygons with white by using an OpenCV filled polygon function;
(4) the vertices of the polygons obtained according to the above process are marked as the water portion of the remote sensing image, the black areas represent the land, and the white areas represent the water.
Further, the image segmentation is to use a merging cost index function for merging the originality, and combine the result of the image annotation.
Further, the spectral average is represented by a calculated average of all pixels constituting an image object in a certain layer:
Further, the area mainly refers to the number of pixels included in the object.
Further, the aspect ratio is an outer rectangular MER (W)MER) Width and MER (L)MER) The ratio of the lengths of (a) to (b) is:
Further, the squareness measures the target by taking the ratio of the area of the target image to the minimum rectangular area around the image as a parameter, that is:
Compared with the prior art, the technical scheme of the application has the following beneficial effects:
1. the regional water resource classification evaluation method based on the improved depth residual error network has the advantages that the image enhancement is to improve the image information and the quality, the characteristics are more obvious under the observation of human eyes, the effect of regional water resource identification is enhanced, the digital interference degree is neglected in the visual analysis, the method plays an important role in various aspects of remote sensing images, the image fusion is to carry out playback sampling on multispectral images with low spatial resolution and single-band images with high spatial resolution to generate new multispectral images with high spatial resolution, so that the generated images have high spatial resolution and multispectral characteristics, a compound mode is mainly adopted to integrate information provided by remote sensing image data sources of different sensors to obtain high-quality image information, meanwhile, information difference among a plurality of sensors is eliminated, the fuzzy degree of the remote sensing image is reduced, and therefore the definition of the image is improved.
2. The regional water resource classification evaluation method based on the improved depth residual error network comprises the steps that after corresponding preprocessing is carried out on remote sensing images through image marking, water resources in the remote sensing images are marked manually, firstly, people need to know which remote sensing images belong to the water resources and which belong to the land, image segmentation is to abstract combined patches into nearest neighbor graphs, the patches are nodes of the nearest neighbor graphs, the adjacency relation of the patches is edges between the nodes, the nodes with the combining cost smaller than the minimum combining cost are found out from any node for combination, a new nearest neighbor graph is generated, if the combining termination condition is not reached, the latest graph is combined again, regional water resource image segmentation is completed, the overall condition of regional water resource distribution in China is analyzed through four steps of image enhancement, image fusion, image marking and image segmentation, and the Gaf-2 satellite remote sensing images are used as the basis of algorithm design, the regional water resource remote sensing image preprocessing method based on the pyramid network comprises the steps of conducting preprocessing such as enhancing, fusing, marking and segmenting on a regional water resource remote sensing image, improving a depth residual error network by using the pyramid network on the basis, combining the improved depth residual error network with a preprocessing result of a general sensing image to obtain a regional water resource classification result, and an experimental result shows that the algorithm has the advantages of being short in preprocessing time, low in classification error rate, short in classification time, good in practical application effect and the like.
Drawings
FIG. 1 is a structural diagram of a Gaussian core of the present invention;
FIG. 2 is a diagram of an IHS modification process of the present invention;
FIG. 3 is a diagram of an original residual error unit according to the present invention;
FIG. 4 is a diagram of the improved depth residual network of the present invention;
FIG. 5 is a graph comparing image pre-processing times according to the present invention;
FIG. 6 is a comparison chart of the classification error rate according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1 to 6, a regional water resource classification evaluation method based on an improved depth residual error network in this embodiment includes four steps of image enhancement, image fusion, image labeling, and image segmentation;
the remote sensing image enhancement aims to improve image information and quality, enable the characteristics to be more obvious under human eye observation, enhance the effect of regional water resource identification, neglect digital interference degree in visual analysis, and play an important role in various aspects of remote sensing images.
Although the domain of the gaussian function covers the whole set of real numbers, in practice, only values within 3 standard deviations of the mean value are needed, and other parts can be directly deleted, wherein the structure diagram of the gaussian core is shown in fig. 1.
The specific steps of carrying out Gaussian low-pass filtering on each wave band of the remote sensing image by using the Gaussian core structure are as follows:
(1) selecting a remote sensing image with obvious water and land resolution, and reading all four wave bands of the image in a gray scale image mode;
(2) a single gaussian filter will blur the edges of the remote sensing image, reducing the contrast, and the two-dimensional gaussian distribution function can be expressed as:
Therefore, the size of the selected Gaussian kernel is 7 x 7, the weight of the middle pixel of the Gaussian kernel is larger, and the weight of the surrounding pixels is smaller, so that the importance of the surrounding pixels can be reduced in the smoothing process, the fuzzy degree of the remote sensing image is reduced, and each wave band of the enhanced remote sensing image is enhanced and de-noised;
(3) directly moving a Gaussian kernel to a pixel to be processed in a remote sensing image;
(4) multiplying the pixel values of the corresponding positions of the remote sensing images by the weight of the Gaussian kernel respectively;
(5) and adding the result obtained in the last step into a pixel value remote sensing image of the area image.
Remote sensing image preprocessing:
when a good secondary satellite generates a remote sensing image, due to the influence of factors such as different or unstable sensor angles, the surface of the earth and the like, the remote sensing image has stretching, distortion and distortion phenomena to a certain extent, the phenomena can cause larger errors generated by using a research result of remote sensing data, the preprocessing of the remote sensing image is to use the existing model or algorithm to restrain or eliminate various influences and errors in image imaging, so that an image which is as real as possible is obtained, in an actual project, different remote sensing image preprocessing methods are selected according to different research purposes and contents, and the chapter preprocesses the remote sensing image from multiple angles to more clearly display the water characteristics of the remote sensing image and improve the accuracy of water resource classification of a final region.
The image fusion is to replay and sample the multispectral image with low spatial resolution and the single-waveband image with high spatial resolution to generate a new multispectral image with high resolution, so that the generated image has the characteristics of high spatial resolution and multispectral, mainly adopts a compound mode, integrates information provided by remote sensing image data sources of different sensors, obtains high-quality image information, eliminates information difference among a plurality of sensors, reduces the fuzzy degree of the remote sensing image, and improves the definition of the image.
Because a single remote sensing image data source has advantages and disadvantages, the fusion of remote sensing images can make the data of different remote sensing data sources complement each other, make up for the deficiency of the single data source, thereby realizing the effective utilization of resources, and has important significance for the extraction of remote sensing image information, especially for the classification and identification of water resources, because the resolution of multispectral data of a high-resolution second-number satellite is 4 meters, the resolution of multispectral data of the high-resolution second-number satellite needs to be improved by adopting an image fusion method, an IHS transformation method is adopted as the fusion method of the remote sensing images, the IHS transformation fusion is mainly color transformation fusion, and is based on the conversion process between an RGB space and an IHS space, the RGB space is a color space mainly synthesized by different wave band remote sensing data obtained by methods such as optics, thermal infrared and radar, and the like, and is a system for describing the color attribute of an object, and the IHS space is a system for extracting the brightness, and the color attribute of the object, The chromaticity space of chromaticity and saturation corresponds to the average radiation intensity, data vector and equivalent data size of each wave band of the image in turn, and after the RGB color space and the HIS chromaticity space are converted, the IHS chromaticity space is inverted to the RGB color space, so that the obtained image has high spatial resolution and multispectral characteristics.
The formula for converting the RGB color space to the IHS chromaticity space is as follows:
wherein,
，
，
the IHS conversion process is shown in fig. 2.
Before determining and classifying regional water resource targets, basic element features of the image markers need to be calculated, and the calculated basic element features comprise a spectrum mean value, an area, a rectangularity and an aspect ratio according to the geometrical features of water resources and the spectrum features of the remote sensing images.
(1) Mean value of spectrum
The spectral average is represented by the calculated average of all pixels constituting an image object in a certain layer:
(2) Area of
The area (a) mainly refers to the number of pixels contained in the object, can be simply implemented using a seed filling algorithm, and can distinguish the size of the object to some extent.
(3) Aspect ratio
Aspect ratio of outer rectangle MER (W)MER) Width and MER (L)MER) The ratio of the lengths of (a) to (b) is:
The aspect ratio can be used to distinguish between elongate objects and square or round objects, in general, the more slender the object the greater the aspect ratio, and the smaller the length and width of a round object or an object shaped close to a square, which has a value of 1.
(4) Degree of rectangularity
The rectangularity of the target is measured by taking the ratio of the area of the target image to the minimum rectangular area around the image as a parameter, namely:
The degree of rectangularity R reflects the degree to which an object fills a closed rectangle, with a maximum value of R for a rectangular object of 1, a minimum value of R for an elongated object of 1, and a value of R for a circular object of pi/4.
After the remote sensing images are correspondingly preprocessed, water resources in the remote sensing images are marked manually, firstly, people need to know which remote sensing images belong to the water resources and which belong to the land, on the basis, an open source image marking tool Labelme on gitubs is utilized, the open source marking tool is modified according to MIT, the water resources of the remote sensing images are marked, Labelmes can construct polygons through tracing points, interested areas are framed out, and polygon coordinates are stored.
(1) Selecting the first three wave bands of the original remote sensing image as three channels of BGR by utilizing the merging function of OpenCV, and synthesizing a new color image presented in a jpg format;
(2) marking the water part in the color image in a polygonal form by using a Labelme image marking tool, and storing the vertexes of the marked polygon during marking;
(3) creating an image with the same size as the original image, defaulting to black, reading the vertex of each polygon, and filling the RGB of pixel points in the polygons with white by using an OpenCV filled polygon function;
(4) the vertices of the polygons obtained according to the above process are marked as the water portion of the remote sensing image, the black areas represent the land, and the white areas represent the water.
The image segmentation is to use a merging cost index function for merging originality, and combine the result of image annotation, wherein the function comprehensively considers the spectral characteristics and the shape characteristics by adopting a weighted average method.
The function is expressed as follows:
Spectral heterogeneity, which is the sum of the differences between the weighted standard deviations of the merged patches and the weighted standard deviations of the merged sub-patches in all bands, can be expressed as:
wherein A is the frequency band weight in the water resource image, c is the frequency band number in the regional water resource image, s represents the standard deviation, n1、n2Is to merge the number of pixels contained in the first two adjacent patches, nmergeIs the number of pixels of the merged patch.
The heterogeneity of shape includes two parts, heterogeneity factor of compactness and heterogeneity factor of smoothness, which can be expressed as:
The heterogeneity coefficient of compactness is expressed by the following formula:
in the formula, nmergeIs the number of pixels of the merged patch, lmergeIs the actual circumference of the merged plaque, n1、n2Is to merge the number of pixels contained in the first two adjacent patches, l1、l2Is the actual perimeter of the merging of the first two adjacent patches.
Firstly, abstracting the graph spots to be merged into a nearest neighbor graph, wherein the graph spots are nodes of the nearest neighbor graph, the adjacent relation of the graph spots is edges between the nodes, the nodes with merging cost smaller than the minimum merging cost are found from any node for merging, a new nearest neighbor graph is generated, and if the merging termination condition is not met, the latest graph is merged again, and the segmentation of the regional water resource image is completed.
Regional water resource classification algorithm based on improved deep residual error network:
the appearance of the deep residual error network enables the image classification field to make a series of breakthroughs, however, the depth of the network is not always the best, more complex functions can be fitted in the network and the network performance is improved along with the increase of the number of network layers in a certain depth range, but when the number of network layers is increased to a certain number, the training precision and the testing precision are rapidly reduced along with the continuous increase of the number of network layers, the network degradation indicates that a deep learning system directly increasing the depth is not easy to be optimized, in order to solve the problem of network degradation caused by the increase of the number of the depth layers, the constant mapping is introduced into the network design under the inspiration of the residual error learning, so that the problems of gradient explosion, gradient disappearance, network degradation and the like caused by the increase of the depth are relieved, the number of information transmission paths is increased, and the network depth is pushed to thousands of layers from tens of layers, the appearance of the deep residual error network greatly improves the precision of the system, and enables the training of the deep residual error network to be feasible, which is a significant breakthrough in the field of image classification.
The depth of the network is directly increased to make the deep learning system difficult to optimize, if a shallow network exists, then a deep network should exist, the deep network is formed by mapping and mapping a plurality of x on the basis of the shallow network, so that the performance of the deep residual error network is at least not worse than that of the shallow network, however, experiments show that the ideal deep residual error network cannot be found, which shows that the constancy mapping of x is difficult to fit by directly stacking depths.
After the shallow network is saturated, an identity mapping layer is added behind the shallow network, so that the depth of the network can be increased, and the error of the system can be ensured not to increase along with the increase of the depth.
If the target to be learned in the residual unit of the original neural network is mapped to h (X), then it may be difficult to learn this target mapping, the deep residual network can make the residual unit not to directly learn the target mapping, but to learn a residual f (X) = h (X) -X, then the original mapping becomes f (X) + X, and the original residual element can be considered to be composed of two parts, a linear mapping X → X and a non-linear mapping f (X), and if X → X is the optimal learning strategy, then it is equivalent to setting the weight parameter of the non-linear mapping f (X) to 0, and the identity mapping makes the non-linear mapping easier to learn the linear X → X mapping.
The residual unit is a basic constituent unit of a depth residual network, and is generally composed of a convolutional layer, a batch normalized Batchnorm layer and a nonlinear activation function, as shown in FIG. 3.
Let the input of the residual unit be XlThen the next layer output is:
wherein, F (X)l,Wl) Is a residual function, WlAre the corresponding weight parameters of the residual function,
a non-linear activation function Relu.
In a depth residual network, the characteristic map size D of k residual units belonging to n groupskAnd as described in the following, the description is given,
in each group of residual block channels, the number of channels is consistent with the size of the characteristic diagram, at the beginning of the next group of residual blocks, the size of the characteristic diagram is halved, and the number of channels is doubled.
This arrangement ensures simplification of data dimensions and increases the diversity of high-level attributes extracted from higher layers of the network by doubling the feature map, but the sudden increase in the number of feature channels makes the transmission of feature information in the network inconsistent, so that some useful information about the prediction is lost and the classification performance of the network is degraded.
Therefore, the pyramid depth residual error network is adopted, the number of channels is gradually increased, the number of channels in the feature map is linearly gradually increased to ensure the diversity of high-level attributes and the continuity of information, the number of channels in the feature map is linearly gradually increased, as shown in the following formula,
wherein,
representing the sum of the number of residual blocks in the improved depth residual network,
representing the number of channels in the profile of the k-residual block output,
representing the step size of the increase in the number of channels.
The improved depth residual network is shown in fig. 4.
The parameters of each layer of the improved depth residual error network are shown in table 1.
TABLE 1 parameters of layers of an improved deep residual network
In order to improve the depth, the residual error network is mainly formed by accumulating the quantity of three groups of residual error blocks at the tail of the same level, the quantity of channels on an output characteristic diagram of the three groups of residual error blocks is gradually increased linearly, and the residual error block D at the tail of each level1Input dimension of =16 channels and the first linear convolution layer of the next residual block is increased by [ a/3 n%]I.e. D2+16+[a/3n]After that each residual block
Channels up to the last residual block output channel number D3nAnd =16+ a, the size of the characteristic diagram is reduced by half step by step from 32 × 32 to 16 × 16 to 8 × 8 at the position of the first residual block in each group, all interlayer connections outside the residual blocks are called interlayer, the interlayer connection outside each residual block group in the first layer is called interlayer second-level connection, and the interlayer connection in the original residual block is called last-level third-level interlayer connection.
The method has the advantages that firstly, the improved depth residual error network inherits the advantages of the depth residual error network, the progressive increase of characteristic dimensionality can enable the information transmission to be smoother, therefore, the information inconsistency caused by the rapid increase of the number of characteristic channels is improved, the structure of a residual error block is used as a core factor of the improved depth residual error network, and the image classification performance of the network can be directly determined.
Inputting the image segmentation result into an improved depth residual error network to obtain a regional water resource image classification result, which can be expressed as:
where K is the size of the low-level features and represents the mutual information between the input and output features computed at each pixel point, and U, V are a super parameter used to weigh the effects of mutual information maximization and prior matching on the training results in the mutual network training process.
And (3) experimental test:
(1) experimental parameters
In order to verify the actual application effect of the regional water resource classification algorithm based on the improved deep residual error network, a simulation experiment test is performed, and the simulation test environment is shown in table 2.
TABLE 2 simulation test Environment project
Taking 1000 remote sensing images as experiment sample data, in order to improve the accuracy of a simulation experiment, on the basis of a reference document [4], a regional water resource classification algorithm based on an artificial fish swarm algorithm and fuzzy C-average clustering, a regional water resource classification algorithm based on a multi-source remote sensing image in a reference document [5], a regional water resource classification algorithm based on a Lagrange's algorithm in a reference document [6] and a regional water resource classification algorithm based on an improved depth residual error network designed in the text are taken as an experiment comparison method, and the comprehensive performance of different algorithms is verified by comparing the image preprocessing time, the classification error rate and the time consumption of the four algorithms.
(2) Analysis of the results of the experiment
Image preprocessing time first, the image preprocessing times of the four algorithms are compared, and as a result, as shown in fig. 5,
from fig. 5, the image preprocessing time of the reference [4] algorithm is as high as 4.7s, the image preprocessing time of the reference [5] algorithm is as high as 4.0s, and the image preprocessing time of the reference [6] algorithm is as high as 4.3s, while the image preprocessing time of the algorithm designed herein is only 0.5s, which shows that the algorithm can quickly realize the image preprocessing of regional water resources.
The classification error rate is compared, the classification error rate of the four algorithms is compared, the comparison result is shown in fig. 6, the comparison result of the classification error of fig. 6 shows that the classification error rate of the algorithm of reference [4] is different from 26% to 33%, the classification error rate of the algorithm of reference [5] is different from 6% to 26%, the classification error rate of the algorithm of reference [6] is different from 6% to 25%, and the classification error rate of the algorithm is set to be less than 3%, so that the algorithm can effectively reduce the classification error of regional water resources and improve the classification accuracy.
The classification takes time, and finally, the classification time of the four algorithms is compared, and the result is shown in table 3.
TABLE 3 comparison of Classification time
Analyzing the data in table 3, the average time of the regional water resource classification algorithm in reference [4] is 2.56s, the average time of the regional water resource classification algorithm in reference [5] is 3.13s, the average time of the regional water resource classification algorithm in reference [6] is 1.62s, the average time of the regional water resource classification algorithm designed herein is 0.20s, which is the lowest time consumption among the four classification algorithms, and thus, the algorithm can rapidly realize regional water resource classification.
And (4) conclusion: the change of the earth environment and the influence thereof on the natural system and the human society are a permanent research topic in the vast scientific field, because of the landform characteristics, climate change and other changes, water resources become one of the most important resources on the earth, a water body is an area defined by clear terrain boundary and water, the condition of the water body in the geographic area is concentrated at specific positions, such as ocean, lake, river or reservoir, in the terrain boundary and the water body boundary, soil, vegetation or a combination of soil and vegetation generally exists, the exploration of the water resource problem is a very important research topic in different fields, such as the management of the coastal area of a lake, the change of a coastline, flood forecasting, water resource evaluation and the like, so the regional water resource classification algorithm based on the improved depth residual error network is proposed herein, the test result shows that the image preprocessing time of the algorithm is short, the regional water resource classification error rate is low, the classification time is short, and the regional water resource can be rapidly and accurately classified.
The working principle of the above embodiment is as follows:
(1) the image enhancement is to improve the image information and quality, make the characteristics more obvious under human eye observation, enhance the effect of regional water resource identification, ignore the digital interference degree in visual analysis, and play an important role in various aspects of remote sensing images.
(2) The image fusion is to carry out playback sampling on a multispectral image with low spatial resolution and a single-band image with high spatial resolution to generate a new multispectral image with high resolution, so that the generated image has high spatial resolution and multispectral characteristics.
(3) After the remote sensing images are correspondingly preprocessed, water resources in the remote sensing images are marked manually, and firstly, people need to know which remote sensing images belong to the water resources and which remote sensing images belong to the land.
(4) And in the image segmentation, the combined image spots are abstracted into a nearest neighbor image, the image spots are nodes of the nearest neighbor image, the adjacency relation of the image spots is edges between the nodes, the nodes with the combining cost smaller than the minimum combining cost are found from any node for combination, a new nearest neighbor image is generated, and if the combining termination condition is not met, the latest image is combined again to complete the regional water resource image segmentation.
(5) The method comprises the steps of image enhancement, image fusion, image marking and image segmentation, the overall situation of regional water resource distribution in China is analyzed, a Gaf-2 satellite remote sensing image is taken as a basis of algorithm design, the regional water resource remote sensing image is subjected to enhancement, fusion, marking, segmentation and other preprocessing, on the basis, a pyramid network is used for improving a depth residual error network, the improved depth residual error network is combined with a preprocessing result of a synaesthesia image, and a regional water resource classification result is obtained.
Reference documents:
[1] chenjinmu, Wang mussel, Wang Xiao Juan, theory the water resource usage control system of China constructs the water conservancy of China, 2017, No.811(01): 23-27.
[2] Libreros J., Bueno G., Trujillo M., Maria O., Automated Identification and Classification of Diatoms from Water Resources[J].River Research & Applications, 2018, 15(2),1544-1552。
[3] German, a discussion on improving the classification system of water resource regions in our country-view [ J ] based on the evolution of the scope of water resource management-proceedings of north china university of water conservancy and hydropower (natural science edition), 2016,37(04): 21-26.
[4] Wanlina, Chengxiang, Liyuean, Linkerong flood classification method based on artificial fish swarm algorithm and fuzzy C-means clustering [ J ] the academic newspaper, 2009,40(06): 743-.
[5] Li Z.X., Lai Z.Q., Long Y.M., Xu G.H., Regional water resources classification algorithm based on multi-source remote sensing image[J].China Rural Water and Hydropower, 2018, 33(11),61-66。
[6] Zhang N.P., Liu Q.H., Regional water resources classification algorithm based on Lagrangian algorithm[J].Henan Water Resources & South-to-North Water Diversion, 2019, 48(10),34-36。
[7] Benevolence, tangyuwen, 2018, construction report of two types of society in Hunan and ecological civilization [ M ]. Beijing: social scientific literature publisher, 2018: 3-5.
[8] Rezaie-Balf M., Zahmatkesh Z., Kim S., Soft Computing Techniques for Rainfall-Runoff Simulation: Local Non–Parametric Paradigm vs. Model Classification Methods[J]. Water Resources Management Dordrecht, 2017, 31(12),3843-3865。
[9] Hargrove W.L., Heyman J.M., A Comprehensive Process for Stakeholder Identification and Engagement in Addressing Wicked Water Resources Problems[J].Land, 2020, 9(4),1-12。
[10] Charles O., Anthony O., Vibrio Pathogens: A Public Health Concern in Rural Water Resources in Sub-Saharan Africa[J]. International Journal of Environmental Research & Public Health, 2017, 14(10),1188-1202。
[11] Gobeyn S., Van Wesemael A., Neal J Lievens H., Van Eerdenbrugh K., De Vleeschouwer N., et al., Impact of the timing of a SAR image acquisition on the calibration of a flood inundation model[J].Advances in Water Resources, 2017, 100(5),126-138。
[12] Zunlei L., Spatial Heterogeneity of Water Resource and Aquaculture Structure in Jiangxi Province Based on Remote Sensing Image[J].Journal of Natural Resources, 2018, 33(10),1833-1846。
[13] Peng Y.T., Cosman P.C., Underwater Image Restoration Based on Image Blurriness and Light Absorption[J], IEEE Trans Image Process, 2017, 26(4),1579-1594。
[14] Akila C., Varatharajan R., Color fidelity and visibility enhancement of underwater image de-hazing by enhanced fuzzy intensification operator[J].Multimedia Tools & Applications, 2017, 77(4),1-14。
[15] Hao Dahai, Yao Yu Zeng, Guanchangqing, pay for construction flight, Liaoning Dalonshan iron ore dump landslide hazard evaluation [ J ] world geology based on high resolution remote sensing technology 2020,39(04): 937-.
[16] Ross M.R.V., Topp S.N., Appling A.P., Yang X., Kuhn C., Butman D., et al., AquaSat: A Data Set to Enable Remote Sensing of Water Quality for Inland Waters[J].Water Resources Research, 2019, 55(11),10012-10025。
[17] Yanlixun, Paixianyu, multi-source remote sensing image data fusion theory and technology [ J ] Beijing surveying and mapping, 2014(03) 101-104.DOI 10.19580/J. cnki.1007-3000.2014.03.025.
[18] Wang S., Li J., Shen Q., Zhang B., Zhang F., Lu Z., et al., MODIS-Based Radiometric Color Extraction and Classification of Inland Water With the Forel-Ule Scale: A Case Study of Lake Taihu[J].IEEE Journal of Selected Topics in Applied Earth Observations & Remote Sensing, 2017, 8(2),907-918。
[19] Mafanya M., Tsele P., Botai J., Manyama P., Swart B., Monate T., et al., Evaluating pixel and object based image classification techniques for mapping plant invasions from UAV derived aerial imagery:Harrisia pomanensis as a case study[J].Isprs Journal of Photogrammetry & Remote Sensing, 2017, 129(6),1-11。
[20] Jódar a J., Carpintero b E., Martos-Rosillo c d S.,et al., Combination of lumped hydrological and remote-sensing models to evaluate water resources in a semi-arid high altitude ungauged watershed of Sierra Nevada (Southern Spain)[J]. Ence of The Total Environment, 2017, 25(1),285-300。
[21] Zhao Zhi Cheng, Luo ze, Wang Peng, Li Jian, research review based on depth residual error network image classification algorithm [ J ]. computer system application, 2020,29(01):14-21.DOI: 10.15888/j.cnki.csa.007243.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (9)
1. A regional water resource classification evaluation method based on an improved depth residual error network is characterized by comprising four steps of image enhancement, image fusion, image marking and image segmentation;
the specific steps of the image enhancement and the Gaussian low-pass filtering are as follows:
(1) selecting a remote sensing image with obvious water and land resolution, and reading all four wave bands of the image in a gray scale image mode;
(2) an independent Gaussian filter blurs the edge of the remote sensing image to reduce the contrast;
(3) directly moving a Gaussian kernel to a pixel to be processed in a remote sensing image;
(4) multiplying the pixel values of the corresponding positions of the remote sensing images by the weight of the Gaussian kernel respectively;
(5) and adding the result obtained in the last step into a pixel value remote sensing image of the area image.
2. The method for classified evaluation of regional water resources based on the improved depth residual error network as claimed in claim 1, wherein the image fusion mainly adopts a compound mode, integrates information provided by remote sensing image data sources of different sensors, obtains high-quality image information, eliminates information differences among multiple sensors, reduces the degree of blur of remote sensing images, and thereby improves the definition of images.
3. The method as claimed in claim 1, wherein the image-labeled basic feature characteristics include a spectral average, an area, a squareness and an aspect ratio.
4. The method for regional water resource classification evaluation based on the improved deep residual error network as claimed in claim 1, wherein the image marking comprises the following steps:
(1) by utilizing the merging function of OpenCV, the first three wave bands of the original remote sensing image are selected as three channels of BGR, and a new color image presented in a jpg format is synthesized;
(2) marking the water part in the color image in a polygonal form by using a Labelme image marking tool, and storing the vertexes of the marked polygon during marking;
(3) creating an image with the same size as the original image, defaulting to black, reading the vertex of each polygon, and filling the RGB of pixel points in the polygons with white by using an OpenCV filled polygon function;
(4) the vertices of the polygons obtained according to the above process are marked as the water portion of the remote sensing image, the black areas represent the land, and the white areas represent the water.
5. The method as claimed in claim 1, wherein the image segmentation is a result of using a merging cost index function for merging originality and combining image labeling.
6. The method as claimed in claim 3, wherein the spectral average is represented by a calculated average of all pixels in a layer that constitute an image object:
7. The method as claimed in claim 3, wherein the area mainly refers to the number of pixels included in the object.
8. The method as claimed in claim 3, wherein the aspect ratio is an outer rectangle MER (W) of the regional water resource classification evaluation method based on the improved depth residual error networkMER) Width and MER (L)MER) The ratio of the lengths of (a) to (b) is:
9. The method as claimed in claim 3, wherein the squareness degree is to measure the target by using the ratio of the area of the target image to the minimum rectangular area around the image as a parameter, namely:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210413863.6A CN114511791A (en) | 2022-04-20 | 2022-04-20 | Regional water resource classification evaluation method based on improved deep residual error network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210413863.6A CN114511791A (en) | 2022-04-20 | 2022-04-20 | Regional water resource classification evaluation method based on improved deep residual error network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114511791A true CN114511791A (en) | 2022-05-17 |
Family
ID=81555326
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210413863.6A Pending CN114511791A (en) | 2022-04-20 | 2022-04-20 | Regional water resource classification evaluation method based on improved deep residual error network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114511791A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107085708A (en) * | 2017-04-20 | 2017-08-22 | 哈尔滨工业大学 | High-resolution remote sensing image change detecting method based on multi-scale division and fusion |
| CN111696123A (en) * | 2020-06-15 | 2020-09-22 | 荆门汇易佳信息科技有限公司 | Remote sensing image water area segmentation and extraction method based on super-pixel classification and identification |
| CN112733745A (en) * | 2021-01-14 | 2021-04-30 | 北京师范大学 | Cultivated land image extraction method and system |
| CN114065831A (en) * | 2021-08-27 | 2022-02-18 | 北京工业大学 | Hyperspectral image classification method based on multi-scale random depth residual error network |
| CN114241297A (en) * | 2021-11-16 | 2022-03-25 | 山东科技大学 | Remote sensing image classification method based on multi-scale pyramid space independent convolution |
-
2022
- 2022-04-20 CN CN202210413863.6A patent/CN114511791A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107085708A (en) * | 2017-04-20 | 2017-08-22 | 哈尔滨工业大学 | High-resolution remote sensing image change detecting method based on multi-scale division and fusion |
| CN111696123A (en) * | 2020-06-15 | 2020-09-22 | 荆门汇易佳信息科技有限公司 | Remote sensing image water area segmentation and extraction method based on super-pixel classification and identification |
| CN112733745A (en) * | 2021-01-14 | 2021-04-30 | 北京师范大学 | Cultivated land image extraction method and system |
| CN114065831A (en) * | 2021-08-27 | 2022-02-18 | 北京工业大学 | Hyperspectral image classification method based on multi-scale random depth residual error network |
| CN114241297A (en) * | 2021-11-16 | 2022-03-25 | 山东科技大学 | Remote sensing image classification method based on multi-scale pyramid space independent convolution |
Non-Patent Citations (3)
| Title |
|---|
| 徐文健: ""基于卷积神经网络的高分辨率遥感图像上的水体识别技术"", 《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》 * |
| 田正杰: ""高分辨率遥感影像道路分割与提取算法研究"", 《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》 * |
| 赵志成 等: ""基于深度残差网络图像分类算法研究综述"", 《计算机系统应用》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cao et al. | A deep learning method for building height estimation using high-resolution multi-view imagery over urban areas: A case study of 42 Chinese cities | |
| CN109840553B (en) | Extraction method and system of cultivated land crop type, storage medium and electronic equipment | |
| CN111598048B (en) | Urban village-in-village identification method integrating high-resolution remote sensing image and street view image | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| CN107491758A (en) | Yangtze river basin water body information and its space encoding method | |
| Xu et al. | Monitoring coastal reclamation changes across Jiangsu Province during 1984–2019 using landsat data | |
| CN107688776B (en) | Urban water body extraction method | |
| WO2024036739A1 (en) | Reservoir water reserve inversion method and apparatus | |
| Poterek et al. | Deep learning for automatic colorization of legacy grayscale aerial photographs | |
| CN105447274A (en) | Method of performing coastal wetland drawing for medium-resolution remote sensing image by utilizing object-oriented classification technology | |
| CN114266958A (en) | Cloud platform based mangrove remote sensing rapid and accurate extraction method | |
| CN110517575A (en) | A kind of surface water body drafting method and device | |
| CN117541930A (en) | River basin scale high space-time resolution river network remote sensing information extraction method and system | |
| Artemeva et al. | Using remote sensing data to create maps of vegetation and relief for natural resource management of a large administrative region | |
| Liu et al. | Spatial-temporal hidden Markov model for land cover classification using multitemporal satellite images | |
| Zhang et al. | Road extraction from multi-source high-resolution remote sensing image using convolutional neural network | |
| Sun et al. | Mapping China’s coastal aquaculture ponds expansion with sentinel-2 images during 2017–2021 | |
| Ma et al. | Urban landscape classification using Chinese advanced high-resolution satellite imagery and an object-oriented multi-variable model | |
| CN116994071A (en) | Multispectral laser radar point cloud classification method based on self-adaptive spectrum residual error | |
| CN117409320A (en) | River basin flood monitoring method, system and storage medium based on satellite remote sensing | |
| CN112966657A (en) | Remote sensing automatic classification method for large-scale water body coverage | |
| Hao et al. | A subpixel mapping method for urban land use by reducing shadow effects | |
| CN114511791A (en) | Regional water resource classification evaluation method based on improved deep residual error network | |
| Guo et al. | Object-Level Hybrid Spatiotemporal Fusion: Reaching a Better Trade-Off Among Spectral Accuracy, Spatial Accuracy and Efficiency | |
| Han et al. | Remote Sensing Image Classification Based on Multi-Spectral Cross-Sensor Super-Resolution Combined With Texture Features: A Case Study in the Liaohe Planting Area |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220517 |