CN116091911A - Automatic identification method and system for buildings in seismic exploration work area - Google Patents
Automatic identification method and system for buildings in seismic exploration work area Download PDFInfo
- Publication number
- CN116091911A CN116091911A CN202111291136.9A CN202111291136A CN116091911A CN 116091911 A CN116091911 A CN 116091911A CN 202111291136 A CN202111291136 A CN 202111291136A CN 116091911 A CN116091911 A CN 116091911A
- Authority
- CN
- China
- Prior art keywords
- building
- work area
- network model
- image
- exploration work
- 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
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention provides an automatic identification method and system for a building in a seismic exploration work area, and belongs to the technical field of intelligent identification of remote sensing images in the seismic exploration work area. The method combines an original Unet network model with a multi-scale convolution characteristic fusion structure and a cavity space pyramid structure to obtain an optimized network model, and utilizes the optimized network model to automatically identify the buildings in the seismic exploration work area. According to the invention, the automatic identification of the remote sensing image building in the seismic exploration work area is realized based on a deep learning method, and the multi-scale convolution feature fusion and cavity space pyramid module ASPP is introduced on the basis of the original Unet network structure, so that the network feature expression capability is enhanced and the semantic information of multiple sizes of targets is obtained. The invention can more accurately position the building target, has more complete target boundary segmentation, has better segmentation effect on small targets and has better universality.
Description
Technical Field
The invention belongs to the technical field of intelligent identification of remote sensing images of a seismic exploration work area, and particularly relates to an automatic identification method and system for buildings in the seismic exploration work area.
Background
The three-dimensional seismic exploration has higher requirements on the seismic data acquisition precision, the coverage area range of a work area is wider, the distribution condition of obstacles represented by buildings in the work area is complex, and the quality of the seismic data acquisition result is directly influenced, so that the operations such as observation system observation and the like are required to be carried out by marking the obstacles such as the buildings before construction. At present, building area information is manually marked based on remote sensing images of a work area in the field of seismic exploration to quickly know building distribution conditions in the work area, but a manual marking mode is low in efficiency, time-consuming and labor-consuming, and cannot meet actual demands.
In recent years, with the rapid development of remote sensing technology, the highest imaging resolution of satellites reaches the sub-meter level, the data volume of high-resolution remote sensing images is rapidly increased, and the use of high-resolution remote sensing images in combination with image processing technology for target identification has become a common method. In the field of field seismic exploration, unmanned aerial vehicles are used for performing the operations of surveying and inspection, and the like, so that convenience is provided for acquiring remote sensing images of exploration work areas. The method for quickly and accurately identifying the building target in the remote sensing image of the seismic exploration work area has great significance for realizing physical point layout, obstacle avoidance, acquisition and other works in the field.
Chinese patent publication CN111652892a discloses a method for extracting and optimizing a building vector of a remote sensing image based on deep learning, which comprises generating a building target probability map from the remote sensing image by using an image segmentation model in a full convolution form, and generating an initial vector contour of the building through binarization, clustering and edge contour tracking processing; the method also comprises the steps of changing the convolution kernel size and improving the PointNet model by applying a recursion filling method, and optimizing the geometric shape of the outline of the building; the method also comprises the step of providing a new loss function to perform PointNet model training, and performing similarity estimation on two polygons with larger node number difference. However, the patent mainly extracts the boundary through the boundary characteristics of the building target, the target characteristics are relatively simple, and the calculation steps are complex.
The remote sensing images of the earthquake exploration work areas have wide distribution range and large quantity of buildings and are difficult to automatically identify. At present, a traditional machine learning method is mainly adopted for extracting building targets in remote sensing images, the method needs to manually set extraction characteristics, subjective consciousness is strong, and recognition efficiency and accuracy are very limited. In recent years, a deep learning method has become a powerful tool for image processing, and therefore, there have been some studies to apply the deep learning method to building segmentation of remote sensing images, but there are problems such as incomplete object boundary segmentation and inaccurate small object segmentation.
Disclosure of Invention
The invention aims to solve the problems in the prior art, and provides an automatic identification method and system for a building in a seismic exploration work area, which are used for quickly and accurately automatically identifying a building target and provide a basis for field physical point layout, obstacle avoidance, acquisition and other works.
The invention is realized by the following technical scheme:
according to a first aspect of the invention, an automatic identification method for a building in a seismic exploration work area is provided, wherein an original Unet network model, a multi-scale convolution characteristic fusion structure and a cavity space pyramid structure are combined to obtain an optimized network model, and the optimized network model is utilized to automatically identify the building in the seismic exploration work area.
The invention further improves that:
the method comprises the following steps:
(1) Collecting an urban remote sensing image dataset related to building target segmentation, and carrying out image preprocessing and dataset division;
(2) Constructing a convolutional neural network;
(3) Constructing a convolutional neural network loss function;
(4) Training the convolutional neural network to obtain an optimized network model;
(5) And automatically identifying the building in the seismic exploration work area by utilizing the optimized network model.
The invention further improves that:
the preprocessing operation in the step (1) comprises the following steps:
(11) Cutting all original images and corresponding tag images into small images with fixed sizes;
(12) Enhancing the original image and the label image;
(13) And compressing the pixel values of the enhanced original image and the label image into a preset value range space to obtain the preprocessed original image and the preprocessed label image.
The data set dividing operation in the step (1) includes:
dividing the preprocessed original image and the preprocessed label image into three data sets according to a set proportion randomly; the three data sets are respectively: experimental training set, validation set and test set.
The invention further improves that:
the operation of step (2) comprises:
the convolutional neural network is constructed as follows: the left side is the coding structure of the original Unet network model, the right side is the decoding structure of the original Unet network model, a cavity space pyramid module ASPP is arranged between the coding structure and the decoding structure, and a multi-scale convolution feature fusion structure is arranged below the coding structure;
the data are simultaneously input into the coding structure and the multi-scale convolution characteristic fusion structure, then the output of each pooling layer of the decoding structure and the output of each convolution layer of the multi-scale convolution characteristic fusion structure are respectively subjected to characteristic splicing operation, and the results obtained by each characteristic splicing operation are respectively input into the decoding structure;
meanwhile, the output result of the last convolution layer of the coding structure is input into the hole space pyramid module ASPP, and the output result of the hole space pyramid module ASPP is input into the decoding structure.
The invention further improves that:
the multi-scale convolution feature fusion structure adopts three convolution kernel sizes, which are respectively: 1*1, 3*3, 5*5.
The invention further improves that:
the space pyramid structure ASPP comprises four space convolution layers, and the space parameters are respectively 1,2,4 and 8.
The operation of step (5) comprises:
preprocessing the acquired data to obtain preprocessed data;
the preprocessed data is input into an optimized network model, the optimized network model outputs a matrix, elements in the matrix are the classification probability of each pixel point, the class probability value of the same pixel point is large, and the automatic identification of the buildings in the seismic exploration work area is realized.
In a second aspect of the invention, an automatic identification system for buildings in a seismic exploration work area is provided:
the system comprises:
the acquisition unit: the method comprises the steps of acquiring an urban remote sensing image dataset related to building target segmentation, and carrying out image preprocessing and dataset division;
a network construction unit: for constructing a convolutional neural network;
function construction unit: the method comprises the steps of constructing a convolutional neural network loss function;
training unit: the system is respectively connected with a network construction unit and a function construction unit and is used for training a convolutional neural network to obtain an optimized network model;
an identification unit: and the training unit is connected with the earthquake exploration work area building automatic identification system and is used for automatically identifying the earthquake exploration work area building by utilizing the optimized network model.
In a third aspect of the present invention, there is provided a computer-readable storage medium storing at least one program executable by a computer, the at least one program when executed by the computer causing the computer to perform the steps in the above-described seismic survey work area building automatic identification method.
Compared with the prior art, the invention has the beneficial effects that:
according to the invention, the automatic identification of the remote sensing image building in the seismic exploration work area is realized based on a deep learning method, and the multi-scale convolution feature fusion and cavity space pyramid module ASPP is introduced on the basis of the original Unet network structure, so that the network feature expression capability is enhanced and the semantic information of multiple sizes of targets is obtained. The invention can more accurately position the building target, has more complete target boundary segmentation, has better segmentation effect on small targets and has better universality. The invention has very important significance for realizing the works of physical point layout, obstacle avoidance, acquisition and the like in the field.
Drawings
FIG. 1 is a block diagram of the steps of the method of the present invention;
FIG. 2 is a network structure diagram of an optimized Unet network model provided by the present invention;
FIG. 3 is a block diagram of a multi-scale convolution feature fusion in accordance with the present invention;
FIG. 4 (a) is an original view of an embodiment of the present invention;
FIG. 4 (b) is a label image corresponding to a model in an embodiment of the present invention;
FIG. 4 (c) shows the result of building segmentation using the method of the present invention in the examples of the present invention.
Detailed Description
The invention is described in further detail below with reference to the attached drawing figures:
aiming at the problems of incomplete target boundary segmentation and inaccurate small target segmentation in the remote sensing image in the prior art, the invention provides a building automatic identification method of a remote sensing image of a seismic exploration work area based on deep learning, which is used for quickly and accurately automatically identifying building targets and provides a basis for field physical point layout, obstacle avoidance, acquisition and other works.
The method comprises the following steps: collecting remote sensing image data for image preprocessing and data set division, constructing an optimized Unet network structure for remote sensing image building segmentation, constructing a network loss function, training an optimized Unet network model by using the remote sensing image data, adjusting network super parameters, testing the network model by using a test data set, and calculating an evaluation index.
The network structure introduces a multi-scale convolution feature fusion and cavity space pyramid module based on a Unet network, so that the network feature expression capability is enhanced and semantic information of more targets is obtained. The remote sensing image data about urban building segmentation is used for training a network model, so that the aims of more complete building target segmentation and more accurate small target segmentation are achieved.
Examples of the method of the invention are as follows:
[ embodiment one ]
As shown in fig. 1, the method specifically comprises the following steps:
s1, acquiring an urban remote sensing image dataset related to building target segmentation, and carrying out image preprocessing and dataset division;
further, the specific steps for the image preprocessing and the data set division in step S1 are as follows:
s11, cutting all original images and corresponding tag images (the two images are open source public data sets, wherein the tag images are black and white binary images, black pixel points represent background, white pixel points represent buildings and are equivalent to marking each pixel point of the original images, the tag images come from the open source public data sets) into small images with fixed sizes by adopting a sliding window mode, and the sliding window size is 1024 x 1024 and the step length is 1024;
s12, enhancing the original image: carrying out image enhancement on an original image by adopting the existing image enhancement methods such as random hue saturation brightness value conversion, random rotation by 90 degrees, random scaling by 15% -25%, random horizontal overturn, random vertical overturn and the like; if the original image is deformed by the data enhancement method, data enhancement is needed to be carried out on the tag image corresponding to the original image, and the tag image is needed to be in one-to-one correspondence with the pixel points of the original image.
S13, performing minimum-maximum standardization treatment on the enhanced original image and the label image, and compressing pixel values of the original image and the label image into a preset value range space to obtain a preprocessed original image and a preprocessed label image;
s14, dividing the preprocessed original image and the preprocessed label image into three data sets, wherein the three data sets are divided randomly according to the ratio of 8:1:1, and the three data sets are respectively: experimental training set, validation set and test set.
S2, constructing a convolutional neural network for remote sensing image building target segmentation;
the invention introduces the existing multi-scale convolution feature fusion structure and the ASPP with the cavity space pyramid structure into the Unet neural network to form a new convolution neural network by combination, and the method comprises the following steps:
an existing Unet network is adopted to form a basic network framework, and the basic network framework comprises an encoding module and a decoding module. The specific steps for constructing the network structure are as follows:
s21, setting an original Unet coding structure, a multi-scale convolution feature fusion structure and a cavity space pyramid module ASPP in the coding module.
The structure organization and data transmission process is as follows:
as shown in fig. 2, the network model of the optimized une of the present invention is centered on ASPP, the left side is the coding structure, the right side is the decoding structure, and the lower side is the multi-scale convolution feature fusion structure.
The original Unet coding structure consists of five convolutions: conv_Block1, conv_Block2, conv_Block3, conv_Block4, conv_Block5, the data transmission is sequentially through these 5 convolutions.
The multi-scale convolution feature fusion structure mainly comprises four multi-scale convolution feature fusion modules, which are respectively: xx1, xx2, xx3, xx4.
The cavity space pyramid structure ASPP mainly comprises four cavity convolution layers.
The decoding structure mainly comprises four deconvolution blocks and four feature fusion modules (the four feature fusion modules are irrelevant to the multi-scale convolution feature fusion module, and the feature fusion module performs a confication operation on a symmetrical coding result, a multi-scale convolution feature fusion result and an output of a corresponding decoding stage), wherein the deconvolution blocks are as follows: deconv_Block1, deconv_Block2, deconv_Block3, deconv_Block4.
The above structures are all existing structures, but the specific use of several multi-scale convolution feature fusion modules, convolution sizes and the size of the cavity space pyramid structure need to be optimized according to specific data and experimental results.
The data is simultaneously input into the coding structure and the multi-scale convolution characteristic fusion structure. In the encoding structure, data passes through conv_block1, maxpooling1, conv_block2, maxpooling2, conv_block3, maxpooling3, conv_block4, maxpooling4, and conv_block5.
In a multi-scale convolution feature fusion structure, data sequentially passes through xx1, xx2, xx3, xx4. And simultaneously, carrying out characteristic splicing operation on Maxpooling1 and xx1, maxpooling2 and xx2, maxpooling3 and xx3 and Maxpooling4 and xx4 respectively (xx 1-xx 4 are output results of the multi-scale convolution characteristic fusion module) so as to carry out characteristic fusion in a subsequent symmetrical decoding structure.
The middle part is an ASPP structure, the output of conv_block5 is input into the ASPP structure, and the output result is input into conv_block6. In the decoding structure, the splicing results of Maxpooling4 and xx4 and the output of Conv_Block6 are spliced, the splicing results of Maxpooling3 and xx3 and the output of Deconv_Block1 are spliced, the splicing results of Maxpooling2 and xx2 and the output of Deconv_Block2 are spliced, the splicing results of Maxpooling1 and xx1 and the output of Deconv_Block3 are spliced, the splicing results of Maxpooling3 and xx3 are input to Deconv Block4 and then the classifying results are output through a softmax layer.
S22, as shown in FIG. 3, three convolution kernel sizes are adopted for each multi-scale convolution feature fusion structure: 1*1, 3*3 and 5*5, performing convolution operation sequentially, selecting four downsampling branches for parallel operation (four downsampling modules in an encoding stage, performing multi-scale convolution feature fusion after each downsampling operation, performing registration operation on a fusion result and a decoding stage feature map, wherein the downsampling operation and the multi-scale convolution feature fusion operation are parallel, and performing registration operation on the parallel result.
S23, the space pyramid structure ASPP comprises four space convolution layers, the space parameters are respectively selected to be 1,2,4 and 8, and finally, the final output of the original Unet coding structure and the output feature graphs of each space convolution layer are added, namely, feature matrixes of outputs of different space parameters are selected to be added.
S24, inputting the data processed in the step S1 into an original Unet coding structure and a multi-scale feature fusion structure at the same time, performing a localization operation (matrix splicing operation) on an output matrix of each pooling layer of the original Unet network and a corresponding multi-scale convolution feature fusion result matrix, performing an operation on the output of Conv_Block5 and the input of Conv_Block5 into a cavity space pyramid module ASPP, and performing up-sampling on the output result. And carrying out the fusion operation of the fusion characteristics of the coding modules Conv_Block4 and xx4 and the deconv_Block1 output characteristic diagram, carrying out the subsequent decoding operation of the fusion results, carrying out the up-sampling operation for 4 times, and finally outputting the building segmentation result.
S3, constructing a loss function of the convolutional neural network;
further, the specific steps for constructing the loss function in step S3 are as follows:
the invention adopts the prior loss function as DiceLoss, and the specific formula is as follows:
in the formula, K represents the total category number in the image, and two targets, namely building pixel points and non-building pixel points, need to be classified, and the value is 2.N represents the total number of pixel points. P is p kn And g kn Are probability values and value rangesIs [0,1]The former is denoted softmax layer (belonging to the decoding structure, the last output layer of the network) output, and the latter is denoted pixel point n predicting the label value belonging to class k.
S4, training the convolutional neural network by using the training set and the verification set to obtain an optimized network model;
and (3) carrying out parameter learning of the network model by combining the training set and the verification set processed in the step (S1) with the network loss function. And (3) through training results, super parameters such as learning rate, training iteration times and the like are adjusted according to an optimization strategy, so that network loss is minimized, and a model with the minimum network loss is selected as a final building segmentation model on the premise that the network is not over-fitted and under-fitted. The step is only required to adopt a conventional network parameter adjusting process, and is not repeated here;
s5, performing network model effect test by using the test data set, outputting a building target extraction and segmentation result, and performing model evaluation IOU index calculation;
further, for step S5, the specific steps of network model test and accuracy evaluation index calculation are as follows:
s51, writing a test code, loading a model, and testing by using a test set to obtain a building segmentation result of the test set;
s52, comparing the test result graph with the label graph, and using the cross-over ratio IOU as an evaluation index, wherein the calculation formula is as follows:
where k+1 represents the pixel class (including a background class), P ij The pixel point with the representative category i is predicted by the model to be the total number of the pixel points with the category j, P ji Representing the total number of pixels with the class j and the class i predicted by the model, P ii The pixel with the category i is predicted by the model to be the total number of pixels with the category i.
When in practical use, the collected data is required to be preprocessed, which specifically comprises the following steps: if the size of the input image is larger, the image is sheared into 1024 x 1024 by using a sliding window method, then the image is subjected to data enhancement, and then the image is input into an optimized network model, the optimized network model outputs a matrix with the size of 1024 x2, the elements in the matrix are the classification probability of each pixel point, and the class probability value of the same pixel point is larger.
The introduction of the multi-scale convolution feature fusion module enables the network to extract feature information of different scales of targets, including small targets, including more target boundary detail features and global information, so that the influence of factors such as shielding illumination on a segmentation result can be eliminated, the anti-interference capability of the network is enhanced, and the building segmentation precision is further improved. The introduction of the cavity space pyramid module enables the network to obtain a larger receptive field, meanwhile, the resolution of the image is not lost too much, and the network extraction features contain more space position information. Therefore, the network related by the invention can achieve the purposes of more complete division of the building target and more accurate division of the small target.
Examples of the application of the method of the invention are as follows:
[ example two ]
The remote sensing image data in a test set is selected for explanation, the implementation flow of the method is shown in figure 1, and the specific steps are as follows:
s1, collecting an urban remote sensing image dataset related to building target segmentation, and carrying out image preprocessing and dataset division;
s2, constructing a convolutional neural network model for remote sensing image building target segmentation;
s3, constructing a convolutional neural network loss function;
s4, training a network model by using the training set and the verification set, and optimizing network super parameters;
s5, performing network model effect test by using the test data set, outputting a building target extraction and segmentation result, and performing model evaluation IOU index calculation;
further, the specific steps for the image preprocessing and the data set division in step S1 are as follows:
s11, cutting all original images and corresponding label images into small images with fixed sizes by adopting a sliding window mode, wherein the sliding window size is 1024 x 1024, and the step length is 1024;
s12, carrying out image enhancement on original remote sensing image data by adopting methods of random hue saturation brightness value transformation, random rotation by 90 degrees, random scaling by 15% -25%, random horizontal overturn, random vertical overturn and the like;
s13, performing minimum-maximum normalization processing on the enhanced data, and compressing pixel values of the original image and the label into a preset value range space;
s14, dividing the preprocessed image and label data into an experimental training set, a verification set and a test set randomly according to the ratio of 8:1:1.
Further, for step S2, the specific steps for constructing the convolutional neural network model for remote sensing image building target segmentation are as follows:
optionally, the network comprises an encoding module and a decoding module. The construction of the network structure is shown in fig. 2, and the specific steps are as follows:
s21, in the encoding module, the encoding module comprises an original Unet encoding structure, a multi-scale convolution feature fusion structure and a cavity space pyramid module ASPP. The original Unet coding structure mainly contains five convolution blocks: conv_Block1, conv_Block2, conv_Block3, conv_Block4, conv_Block5. The multi-scale convolution feature fusion structure mainly comprises four multi-scale convolution feature fusion modules: xx1, xx2, xx3, xx4 (four multi-scale convolution feature fusion modules, xx1, xx2, xx3, xx4 in the order of left to right in fig. 2). The spatial pyramid structure ASPP mainly comprises four spatial convolution layers. The decoder mainly comprises four deconvolution blocks and four feature fusion modules, wherein the deconvolution blocks are as follows: deconv_Block1, deconv_Block2, deconv_Block3, deconv_Block4.
S22, selecting three convolution kernel sizes in the multi-scale convolution feature fusion structure: 1*1, 3*3 and 5*5, performing convolution operation sequentially, selecting four downsampling branches for parallel operation, and performing the localization operation on the parallel results, wherein the multi-scale convolution characteristic fusion network structure is shown in fig. 3.
S23, the space pyramid structure ASPP comprises four space convolution layers, the space parameters are respectively selected to be 1,2,4 and 8, and finally, the final output of the original Unet coding structure and the output feature diagram of each space convolution layer are added.
S24, inputting the data processed in the step S1 into an original Unet coding structure and a multi-scale feature fusion structure at the same time, performing a localization operation on the input result of each pooling layer and the corresponding multi-scale convolution feature fusion result, performing an operation on the output of Conv_Block5 and the input of Conv_Block5 into a cavity space pyramid module ASPP, and performing up-sampling on the output result. And carrying out the fusion operation of the fusion characteristics of the coding modules Conv_Block4 and xx4 and the deconv_Block1 output characteristic diagram, carrying out the subsequent decoding operation of the fusion results, carrying out the up-sampling operation for 4 times, and finally outputting the building segmentation result.
Further, the specific steps for constructing the loss function in step S3 are as follows:
the loss function adopted by the invention is DiceLoss, and the specific formula is as follows:
in the formula, K represents the total category number in the image, and two targets, namely building pixel points and non-building pixel points, need to be classified, and the value is 2.N represents the total number of pixel points. P is p kn And g kn Are all probability values with the value range of 0,1]The former is denoted softmax layer output and the latter is denoted pixel point n predicting the label value belonging to class k.
Further, the specific steps for training the network model in step S4 are as follows:
and (3) carrying out parameter learning of the network model by combining the training set and the verification set processed in the step (S1) with the network loss function. According to training results, super parameters such as learning rate, training iteration times and the like are adjusted according to an optimization strategy, so that network loss is minimized, and a model with the minimum network loss is selected as a final building segmentation model on the premise that the network is not fitted and under-fitted;
further, for step S5, the specific steps of network model test and accuracy evaluation index calculation are as follows:
s51, writing a test code, loading a model, and testing by using a test set, wherein an original image shown in fig. 4 (a) is selected for model testing, a corresponding label image is shown in fig. 4 (b), and a building segmentation result is obtained, as shown in fig. 4 (c);
s52, comparing and calculating the segmentation result graph and the label graph of the test set, and using the cross-over ratio IOU as an evaluation index, wherein the calculation formula is as follows:
where k+1 represents the pixel class (including a background class), P ij The pixel point with the representative category i is predicted by the model to be the total number of the pixel points with the category j, P ji Representing the total number of pixels with the class j and the class i predicted by the model, P ii The pixel with the category i is predicted by the model to be the total number of pixels with the category i. As shown in table 1, the present invention can be used to extract the whole image feature more completely.
| Detection method | IoU/％ |
| Unet | 68.14 |
| The method of the invention | 75.69 |
TABLE 1
The invention also provides an automatic identification system for the buildings in the seismic exploration work area, and the embodiment of the system is as follows:
[ example III ]
The system comprises:
the acquisition unit: the method comprises the steps of acquiring an urban remote sensing image dataset related to building target segmentation, and carrying out image preprocessing and dataset division;
a network construction unit: for constructing a convolutional neural network;
function construction unit: the method comprises the steps of constructing a convolutional neural network loss function;
training unit: the system is respectively connected with a network construction unit and a function construction unit and is used for training a convolutional neural network to obtain an optimized network model;
an identification unit: and the training unit is connected with the earthquake exploration work area building automatic identification system and is used for automatically identifying the earthquake exploration work area building by utilizing the optimized network model.
The present invention also provides a computer-readable storage medium, an embodiment of which is as follows:
[ example IV ]
The computer-readable storage medium stores at least one program executable by a computer, which when executed by the computer, causes the computer to perform the steps in the above-described seismic survey work area building automatic identification method.
The invention can extract the whole image feature, and completely divide the target boundary by optimizing the Unet, and introduces multi-scale convolution feature fusion on the basis of increasing the cavity space pyramid module ASPP, so that the network width is expanded, the network can extract the features of different sizes of the target, and the feature extraction capability of the network is enhanced.
In summary, the invention introduces the multi-scale convolution feature fusion and the cavity space pyramid module based on the Unet network structure, thereby enhancing the network feature expression capability and obtaining semantic information of more targets. The remote sensing image data about urban building segmentation is used for training a network model, so that the aims of more complete building target segmentation and more accurate small target segmentation are achieved. The invention provides an effective method for rapid and high-precision automatic identification of the building of the remote sensing image of the seismic exploration work area, greatly improves the related working efficiency, and has important significance for the subsequent seismic exploration work.
In the description of the present invention, it should be noted that, unless explicitly specified and limited otherwise, the terms "connected," "connected," and "connected" are to be construed broadly, and may be either fixedly connected, detachably connected, or integrally connected, for example; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
In the description of the present invention, unless otherwise indicated, the terms "upper," "lower," "left," "right," "inner," "outer," and the like are used for convenience in describing the present invention and simplifying the description based on the orientation or positional relationship shown in the drawings, and do not denote or imply that the devices or elements in question must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention.
Finally, it should be noted that the above-mentioned technical solution is only one embodiment of the present invention, and various modifications and variations can be easily made by those skilled in the art based on the application methods and principles disclosed in the present invention, and are not limited to the methods described in the above-mentioned specific embodiments of the present invention, therefore, the foregoing description is only preferred, and not meant to be limiting.
Claims (10)
1. An automatic identification method for buildings in a seismic exploration work area is characterized by comprising the following steps of: the method combines an original Unet network model with a multi-scale convolution characteristic fusion structure and a cavity space pyramid structure to obtain an optimized network model, and utilizes the optimized network model to automatically identify the buildings in the seismic exploration work area.
2. The method for automatically identifying a building in a seismic survey work area of claim 1, wherein: the method comprises the following steps:
(1) Collecting an urban remote sensing image dataset related to building target segmentation, and carrying out image preprocessing and dataset division;
(2) Constructing a convolutional neural network;
(3) Constructing a convolutional neural network loss function;
(4) Training the convolutional neural network to obtain an optimized network model;
(5) And automatically identifying the building in the seismic exploration work area by utilizing the optimized network model.
3. The method for automatically identifying a building in a seismic survey work area of claim 2, wherein: the operation of preprocessing in the step (1) comprises the following steps:
(11) Cutting all original images and corresponding tag images into small images with fixed sizes;
(12) Enhancing the original image and the label image;
(13) And compressing the pixel values of the enhanced original image and the label image into a preset value range space to obtain the preprocessed original image and the preprocessed label image.
4. A method for automatically identifying a building in a seismic survey work area according to claim 3, wherein: the data set dividing operation in the step (1) includes:
dividing the preprocessed original image and the preprocessed label image into three data sets according to a set proportion randomly; the three data sets are respectively: experimental training set, validation set and test set.
5. The method for automatically identifying a building in a seismic survey work area of claim 1, wherein: the operation of step (2) comprises:
the convolutional neural network is constructed as follows: the left side is the coding structure of the original Unet network model, the right side is the decoding structure of the original Unet network model, a cavity space pyramid module ASPP is arranged between the coding structure and the decoding structure, and a multi-scale convolution feature fusion structure is arranged below the coding structure;
the data are simultaneously input into the coding structure and the multi-scale convolution characteristic fusion structure, then the output of each pooling layer of the decoding structure and the output of each convolution layer of the multi-scale convolution characteristic fusion structure are respectively subjected to characteristic splicing operation, and the results obtained by each characteristic splicing operation are respectively input into the decoding structure;
meanwhile, the output result of the last convolution layer of the coding structure is input into the hole space pyramid module ASPP, and the output result of the hole space pyramid module ASPP is input into the decoding structure.
6. The method for automatically identifying a building in a seismic survey work area of claim 5, wherein: the multi-scale convolution feature fusion structure adopts three convolution kernel sizes, which are respectively: 1*1, 3*3, 5*5.
7. The method for automatically identifying a building in a seismic survey work area of claim 5, wherein: the space pyramid structure ASPP comprises four space convolution layers, and the space parameters are respectively 1,2,4 and 8.
8. The method for automatically identifying a building in a seismic survey work area of claim 2, wherein: the operation of step (5) comprises:
preprocessing the acquired data to obtain preprocessed data;
the preprocessed data is input into an optimized network model, the optimized network model outputs a matrix, elements in the matrix are the classification probability of each pixel point, the class probability value of the same pixel point is large, and the automatic identification of the buildings in the seismic exploration work area is realized.
9. An automatic identification system for buildings in a seismic exploration work area is characterized in that: the system comprises:
the acquisition unit: the method comprises the steps of acquiring an urban remote sensing image dataset related to building target segmentation, and carrying out image preprocessing and dataset division;
a network construction unit: for constructing a convolutional neural network;
function construction unit: the method comprises the steps of constructing a convolutional neural network loss function;
training unit: the function building unit is respectively connected with the network building unit and the function building unit and is used for training the convolutional neural network to obtain an optimized network model;
an identification unit: and the training unit is connected with the earthquake exploration work area building automatic identification system and is used for automatically identifying the earthquake exploration work area building by utilizing the optimized network model.
10. A computer-readable storage medium, characterized by: the computer-readable storage medium stores at least one program executable by a computer, which when executed by the computer, causes the computer to perform the steps in the seismic survey work area building automatic identification method of any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111291136.9A CN116091911A (en) | 2021-10-29 | 2021-10-29 | Automatic identification method and system for buildings in seismic exploration work area |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111291136.9A CN116091911A (en) | 2021-10-29 | 2021-10-29 | Automatic identification method and system for buildings in seismic exploration work area |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116091911A true CN116091911A (en) | 2023-05-09 |
Family
ID=86210647
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111291136.9A Pending CN116091911A (en) | 2021-10-29 | 2021-10-29 | Automatic identification method and system for buildings in seismic exploration work area |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116091911A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116433991A (en) * | 2023-06-14 | 2023-07-14 | 中国地质大学（武汉） | Post-earthquake building damage assessment method for emergency rescue |
-
2021
- 2021-10-29 CN CN202111291136.9A patent/CN116091911A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116433991A (en) * | 2023-06-14 | 2023-07-14 | 中国地质大学（武汉） | Post-earthquake building damage assessment method for emergency rescue |
| CN116433991B (en) * | 2023-06-14 | 2023-08-22 | 中国地质大学（武汉） | Post-earthquake building damage assessment method for emergency rescue |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109446992B (en) | Remote sensing image building extraction method and system based on deep learning, storage medium and electronic equipment | |
| CN110136170B (en) | Remote sensing image building change detection method based on convolutional neural network | |
| CN103400151B (en) | The optical remote sensing image of integration and GIS autoregistration and Clean water withdraw method | |
| CN110163213B (en) | Remote sensing image segmentation method based on disparity map and multi-scale depth network model | |
| EP3345129A1 (en) | Systems and methods for analyzing remote sensing imagery | |
| CN112381013B (en) | Urban vegetation inversion method and system based on high-resolution remote sensing image | |
| US20220004740A1 (en) | Apparatus and Method For Three-Dimensional Object Recognition | |
| CN114398948A (en) | Multispectral image change detection method based on space-spectrum combined attention network | |
| CN114612883A (en) | Forward vehicle distance detection method based on cascade SSD and monocular depth estimation | |
| CN111709387A (en) | Building segmentation method and system for high-resolution remote sensing image | |
| CN116524189A (en) | High-resolution remote sensing image semantic segmentation method based on coding and decoding indexing edge characterization | |
| CN114358133B (en) | Method for detecting looped frames based on semantic-assisted binocular vision SLAM | |
| CN113033386B (en) | High-resolution remote sensing image-based transmission line channel hidden danger identification method and system | |
| CN116091911A (en) | Automatic identification method and system for buildings in seismic exploration work area | |
| CN112784869B (en) | Fine-grained image identification method based on attention perception and counterstudy | |
| CN110929739B (en) | Automatic impervious surface range remote sensing iterative extraction method | |
| CN116664954A (en) | Hyperspectral ground object classification method based on graph convolution and convolution fusion | |
| CN113850769A (en) | Hyperspectral change detection method based on Simese space spectrum joint convolution network | |
| CN112069445A (en) | 2D SLAM algorithm evaluation and quantification method | |
| Wang et al. | A Cascaded Cross-Modal Network for Semantic Segmentation from High-Resolution Aerial Imagery and RAW Lidar Data | |
| CN117115566B (en) | Urban functional area identification method and system by utilizing full-season remote sensing images | |
| CN114581771B (en) | Method for detecting collapse building by high-resolution heterogeneous remote sensing | |
| Liu et al. | Identification of Damaged Building Regions from High-Resolution Images Using Superpixel-Based Gradient and Autocorrelation Analysis | |
| CN117853914A (en) | Remote sensing image building identification method based on deep learning and space object | |
| Yuan et al. | Fusion Network for Building Extraction Using High-Resolution Aerial Images and LiDAR Data. Remote Sens. 2021, 13, 2473 |
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 |