CN109740485B - Reservoir or small reservoir identification method based on spectral analysis and deep convolutional neural network - Google Patents

Reservoir or small reservoir identification method based on spectral analysis and deep convolutional neural network Download PDF

Info

Publication number
CN109740485B
CN109740485B CN201811607578.8A CN201811607578A CN109740485B CN 109740485 B CN109740485 B CN 109740485B CN 201811607578 A CN201811607578 A CN 201811607578A CN 109740485 B CN109740485 B CN 109740485B
Authority
CN
China
Prior art keywords
neural network
convolutional neural
water
reservoir
spectral analysis
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201811607578.8A
Other languages
Chinese (zh)
Other versions
CN109740485A (en
Inventor
杨明祥
冯涛
王浩
雷晓辉
蒋云钟
权锦
董宁澎
林锋
余琅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Institute of Water Resources and Hydropower Research
Original Assignee
China Institute of Water Resources and Hydropower Research
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by China Institute of Water Resources and Hydropower Research filed Critical China Institute of Water Resources and Hydropower Research
Priority to CN201811607578.8A priority Critical patent/CN109740485B/en
Publication of CN109740485A publication Critical patent/CN109740485A/en
Application granted granted Critical
Publication of CN109740485B publication Critical patent/CN109740485B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Analysis (AREA)

Abstract

The invention discloses a reservoir or small reservoir identification method based on spectral analysis and a deep convolutional neural network, and relates to the field of artificial intelligence and image identification. The method comprises the steps of firstly processing a high-definition remote sensing image by utilizing spectral analysis, extracting water area boundary information, mapping the water area boundary information to an original image, then dividing an original image into a plurality of image blocks containing water storage bodies by utilizing the water area boundary information on the original image, taking the image blocks as the input of a constructed convolutional neural network, identifying and classifying dams, gates and spillways by utilizing the convolutional neural network, and finally identifying and classifying the water bodies by utilizing the dams, gates and spillways. The accuracy rate reaches more than 80%, the accuracy of water body identification is greatly improved, the convolutional neural network is obviously simplified in structure, the running speed and the robustness are greatly improved, the processing speed can reach 23fps, and the rapid and accurate identification of reservoirs and small reservoirs is realized.

Description

Reservoir or small reservoir identification method based on spectral analysis and deep convolutional neural network
Technical Field
The invention relates to the field of artificial intelligence and image recognition, in particular to a reservoir or small reservoir recognition method based on spectral analysis and a deep convolutional neural network.
Background
The medium and small rivers are widely distributed, the river network is dense, the storage and dredging capacity for sudden precipitation is weak, and reservoirs and small dams distributed in medium and small river basins have influence on the production convergence of the river basins through different flood storage and discharge mechanisms. In flood forecasting of medium and small rivers, reservoirs and small dams have important influence on the flood process of a river basin, and the reservoirs and the small dams need to be considered respectively because of different flood storage and discharge mechanisms.
The identification and classification of the reservoir and the small and medium river dams are important links for flood forecasting of the medium and small rivers, the reservoir and the small and medium river dams need to be identified and considered respectively, and data support is provided for flood forecasting of the medium and small rivers through corresponding digital elevation and high-definition remote sensing images and different flood storage and discharge mechanisms of the reservoir and the small and medium river dams.
At present, the identification and classification of reservoirs and small reservoirs mainly adopts other methods based on deep learning. Such methods generally use a region-based convolutional neural network (R-CNN) to generate a suggestion box (region pro-posal), classify features in the suggestion box, and perform regression on the suggestion box to obtain the location and the category of the target. The method has the problems that the network structure is too complex, the requirements on computer hardware conditions are high, the running speed is low, and professional remote sensing practitioners are needed when the training set is labeled, so that the method is greatly limited in the identification process.
Disclosure of Invention
The invention aims to provide a reservoir or small reservoir identification method based on spectral analysis and a deep convolutional neural network, so as to solve the problems in the prior art.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a reservoir or small reservoir identification method based on spectral analysis and a deep convolutional neural network comprises the following steps:
s1, constructing a convolutional neural network, wherein the network sequentially comprises an input layer, five convolutional layers, a fusion layer, a full-connection layer and an output layer, and the fusion layer is used for fusing the characteristics of the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer;
s2, analyzing the high-resolution satellite remote sensing image by adopting a multi-band spectral relation method based on a threshold value, extracting water area boundary information, framing a water body, and obtaining an image after spectral analysis processing;
s3, mapping the coordinates of the original image and the image after spectral analysis to the same coordinate system, mapping the water area boundary information to the original image under the same coordinate system, and realizing framing of the water body on the original image;
s4, dividing the original picture into a plurality of picture blocks containing water storage bodies by using the water area boundary information on the original image, and taking the picture blocks as the input of a convolutional neural network;
s5, extracting the features of the input picture block containing the water storage body by using the convolutional neural network, and extracting the feature in the input picture block: identifying a dam, a gate or a spillway;
and S6, further classifying the input picture blocks according to the output result of the convolutional neural network, namely whether a dam, a gate and a spillway exist, determining the specific type of the water body, namely a reservoir or a small reservoir, and combining the water body boundary information to realize the identification of the water body.
Preferably, the step of S1 further comprises: and training the convolutional neural network by using a training set, and evaluating the accuracy of the trained convolutional neural network by using a testing set.
Preferably, the training set and the test set are high-definition image data sets including dams, gates and spillways, the size of each image is 224 × 224 pixels, the images are ordinary grayscale images, the ratio of the training set to the test set is 9:1, and the maximum number of training iteration steps is set to be 100K.
Preferably, the training of the convolutional neural network is performed by using a training set, specifically: the number of pictures used for training at a time is 128, i.e., min _ batch is 128; the dynamic learning rate is adopted, when the number of iteration steps is less than 30K, the value of the learning rate is 0.0001, when the number of iteration steps reaches 30K, the value of the learning rate is 0.00002, and when the number of iteration steps reaches 80K, the value of the learning rate is 0.00001.
Preferably, the evaluation criteria are recall, precision, F1 and mapp, respectively calculated according to the following formulas:
Figure BDA0001923934930000031
Figure BDA0001923934930000032
Figure BDA0001923934930000033
Figure BDA0001923934930000034
in the formulas (1) and (2), Truepositive represents a positive sample predicted to be positive by the model, namely the accuracy rate of true judgment; falseneractive represents the positive samples predicted to be negative by the model, i.e. the rate of missing reports; falsepositiotive represents a negative sample predicted to be positive by the model, namely a false alarm rate;
in the formula (4), N is the number of all pictures in the test set, M is the total number of target categories, p (k) represents the value of precision when k pictures can be identified, and Δ r (k) represents the change of recall when the number of identified pictures changes from k-1 to k.
Preferably, S2 specifically includes analyzing the high-resolution satellite remote sensing image by using the following formula:
(b3+b4)-(b5+b6)>T (5)
in the formula, b3 is the gray value of the wave band 3; b4 is the gray value of band 4; b5 is the gray value of wave band 5; b6 is the gray value of wave band 6; t is a threshold value selected through experiments;
the band is calculated by using the formula, and the water body is obtained when the requirement is met.
The invention has the beneficial effects that: the invention provides a reservoir or small reservoir identification method based on spectral analysis and a deep convolutional neural network, which comprises the steps of firstly processing a high-definition remote sensing image by utilizing the spectral analysis, extracting water area boundary information, mapping the water area boundary information onto an original image, then dividing an original image into a plurality of picture blocks containing a water storage body by utilizing the water area boundary information on the original image, taking the picture blocks as the input of a constructed convolutional neural network, identifying and classifying a dam, a gate and a spillway by utilizing the convolutional neural network, and finally identifying and classifying the water bodies (the reservoir and the small reservoir) by utilizing the dam, the gate and the spillway. Therefore, the method provided by the invention has the advantages that the accuracy rate reaches more than 80%, the accuracy of water body identification is greatly improved, the structure of the convolutional neural network is obviously simplified, the running speed and the robustness are greatly improved, the processing speed can reach 23fps, and the rapid and accurate identification of reservoirs and small reservoirs is realized.
Drawings
FIG. 1 is a schematic flow chart of a reservoir or small reservoir dam identification method based on spectral analysis and a deep convolutional neural network provided by the invention;
FIG. 2 is a schematic diagram of a VGG-16 network architecture;
FIG. 3 is a schematic diagram of a convolutional neural network constructed according to the present invention;
FIG. 4 is a schematic diagram of the principle of using a test set to evaluate the accuracy of a convolutional neural network;
FIG. 5 is an original satellite picture;
FIG. 6 is a picture processed by a threshold-based inter-multi-band spectral relationship;
FIG. 7 is a schematic diagram of a convolutional neural network structure constructed by the present invention;
FIG. 8 is a photograph of three types of ground features including dams, gates, and spillways;
FIG. 9 is a schematic diagram of a file structure of a data set;
FIG. 10 is a high-definition satellite image of a reservoir in a research area;
FIG. 11 is a high definition satellite image of a study plot of a small reservoir;
FIG. 12 is an electromagnetic spectrum;
FIG. 13 is a photograph of the reservoir after spectral analysis;
fig. 14 is an image of the pond dam after spectral analysis.
In fig. 7, input data of the neural network is shown, and the input data is a high-resolution remote sensing image. Representing the input layer of the convolutional neural network, and the structure in the square frame is five convolutional layers and is used for extracting the characteristics of input data. And (iv) a feature fusion layer for fusing the feature of the second convolutional layer (ii), (iii), nini, and (v). The layer is a full connection layer, the layer adopts a full connection mode, and each neuron is connected with other neurons. And sixthly, outputting the classification result of the neural network to an output layer.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
As shown in fig. 1, the present invention provides a reservoir or small reservoir identification method based on spectral analysis and deep convolutional neural network, comprising the following steps:
s1, constructing a convolutional neural network, wherein the network sequentially comprises an input layer, five convolutional layers, a fusion layer, a full-connection layer and an output layer, and the fusion layer is used for fusing the characteristics of the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer;
s2, analyzing the high-resolution satellite remote sensing image by adopting a multi-band spectral relation method based on a threshold value, extracting water area boundary information, framing a water body, and obtaining an image after spectral analysis processing;
s3, mapping the coordinates of the original image and the image after spectral analysis to the same coordinate system, mapping the water area boundary information to the original image under the same coordinate system, and realizing framing of the water body on the original image;
s4, dividing the original picture into a plurality of picture blocks containing water storage bodies by using the water area boundary information on the original image, and taking the picture blocks as the input of a convolutional neural network;
s5, extracting the features of the input picture block containing the water storage body by using the convolutional neural network, and extracting the feature in the input picture block: identifying a dam, a gate or a spillway;
and S6, further classifying the input picture blocks according to the output result of the convolutional neural network, namely whether a dam, a gate and a spillway exist, determining the specific type of the water body, namely a reservoir or a small reservoir, and combining the water body boundary information to realize the identification of the water body.
Because the remote sensing image ground object size is less, receives the influence of shooting angle and illumination condition great, along with the deepening of network, the characteristic of extraction is more and more abstract, and the size of picture also is more and more littleer, adopts traditional network structure can lose the information of small-size ground object in the later stage. Therefore, a reasonable network structure needs to be designed. The method for solving the problem of identifying the small ground objects can adopt a method of propagating and fusing the characteristics of a plurality of convolution layers backwards.
Experiments have shown that the combination of several small filter convolutional layers is better than one large filter convolutional layer. The VGG-16 network makes full use of the structure of a plurality of convolution layers and nonlinear active layers which are alternated, and can extract deep features compared with the structure of a single convolution layer. The VGG-16 (the network structure of VGG-16 is shown in FIG. 2) network comprises 5 convolutional layers and 3 fully-connected layers, each convolutional layer is composed of a plurality of sub-layers, and the sizes of convolution kernels used are 3 x 3. The feature map numbers output by the 5 convolutional layers are respectively 64,128,256,512 and 512, the features are expanded into 4096-dimensional vectors in the first fully-connected layer, and a 3-dimensional vector is obtained in the third fully-connected layer. And outputting the classification result at an output layer.
In the invention, a VGG-16 network is used as a basic structure to construct a convolutional neural network, and the features extracted from the second, third, fourth and fifth convolutional layers are fused on the basis of the VGG-16 network and used as the input of a full connection layer, as shown in FIG. 3, so that the classification precision of small objects is improved.
After the convolutional neural network is constructed, the convolutional neural network may be trained using a training set, and the accuracy of the trained convolutional neural network may be evaluated using a test set.
During training and testing, the used data set is a high-definition image data set comprising dams, gates and spillways, and the size of each picture is 224 multiplied by 224 pixels and is a common gray picture. The number of pictures contained in the data set was 60000, and the ratio of the training set to the test set was 9:1, i.e., the test set contained 54000 pictures and the test set contained 6000 pictures. And stopping training when the iteration step number reaches 100K, and storing the trained model for subsequent water storage body identification.
Training the convolutional neural network by using a training set, specifically: the number of pictures used for training at a time is 128; the dynamic learning rate is adopted, when the number of iteration steps is less than 30K, the value of the learning rate is 0.0001, when the number of iteration steps reaches 30K, the value of the learning rate is 0.00002, and when the number of iteration steps reaches 80K, the value of the learning rate is 0.00001.
For the trained neural network, the accuracy of the network is evaluated by using a test set (the test set comprises 6000 pictures), and the evaluation criteria are recall, precision, F1 and mAP. As shown in fig. 4 and formulas (1), (2), (3) and (4).
Figure BDA0001923934930000071
Figure BDA0001923934930000072
Figure BDA0001923934930000073
Figure BDA0001923934930000074
In the formulas (1) and (2), True positive represents a positive sample predicted to be positive by the model, namely the True accuracy; false negative represents a positive sample predicted to be negative by the model, namely a False negative rate; false positive represents a negative sample predicted to be positive by the model, namely a False alarm rate;
in the formula (4), N is the number of all pictures in the test set, M is the total number of target categories, p (k) represents the value of precision when k pictures can be identified, and Δ r (k) represents the change of recall when the number of identified pictures changes from k-1 to k.
In spectroscopic analysis, the spectroscopic characteristics of a water body are: the reflection of the water body is mainly concentrated in a blue-green wave band in a visible light wave band, and the reflectivity of the water body begins to be rapidly attenuated to 0 in a near infrared wave band. And the reflectivity of other ground objects such as vegetation and soil is obviously higher than that of water bodies in the near infrared band. Therefore, on the near-infrared remote sensing image, the water body generally presents dark tone due to low reflectivity, vegetation, soil and the like generally present bright tone due to the reflectivity far higher than that of the water body, and the water body can be distinguished from other ground objects through the contrast of the two.
The electromagnetic spectrum is shown in fig. 12, wherein 400nm-700nm is visible light band, 750nm and 800nm are near infrared band, the reflectivity of water body is obviously attenuated at 750nm band, and the reflectivity of water body is attenuated to 0 at 800nm band.
The method selects a threshold-based multi-band inter-spectrum relation method to analyze the high-definition satellite remote sensing image in the research area. The Landsat 8 OLI image contains 9 wave bands, wherein b3 is a green light wave band, b4 is a red light wave band, b5 is a near infrared wave band, b6 is a short wave infrared wave band, and the unique inter-spectral feature of the water body is that the sum of b3 gray value and b4 gray value is larger than the sum of b5 gray value and b6 gray value, so that the water body can be well extracted by utilizing the inter-spectral relationship.
S2 specifically, the high-resolution satellite remote sensing image is analyzed by using the following formula:
(b3+b4)-(b5+b6)>T (5)
in the formula, b3 is the gray value of the wave band 3; b4 is the gray value of band 4; b5 is the gray value of wave band 5; b6 is the gray value of wave band 6; t is a threshold value selected by experiments, and the value of T is 400 by comparison.
And (5) calculating the wave band by using the formula (5), wherein the water body meeting the formula (5) is obtained.
In the actual use process, the accuracy of spectral analysis can be evaluated, the input data is a high-resolution remote sensing image, and the evaluation indexes are the proportion rho of the number of the water bodies detected by the spectral analysis to the actual number of the water bodies and the accuracy phi determined by a water body boundary frame. Setting the actual water quantity as delta and the detected water quantity as epsilon, and defining the detection ratio rho as shown in formula (6):
Figure BDA0001923934930000091
the detection ratio p should be greater than 90%.
Accuracy of water area bounding box determination
Figure BDA0001923934930000098
Is an index for measuring the extraction accuracy of the boundary information of the water area. Let (x, y) denote the coordinates of the upper left corner of the actual water bounding box α, and h, w denote the width and height of α, respectively. (x1, y1) represents the coordinates of the upper left corner of the spectral analysis water bounding box β, and h1, w1 represent the width and height of β, respectively. The transformation from β to α can be represented by equations (7) (8) (9) (10):
Figure BDA0001923934930000092
Figure BDA0001923934930000093
Figure BDA0001923934930000094
Figure BDA0001923934930000095
accuracy of boundary framing in water area
Figure BDA0001923934930000096
Is as defined in formula (11):
Figure BDA0001923934930000097
in the invention, on the basis of extracting the water body by using spectral analysis and classifying the ground objects by using a training network, the input of unknown classes is identified, and information such as the class position of the water body is output. And identifying the dam, the gate and the spillway by using the convolutional neural network, and further determining the category of the impounded body by identifying the dam, the gate and the spillway.
And fusing the water body position information obtained by the spectral analysis and the water body category information obtained by the convolutional neural network to determine the category and the position of the water storage body and realize the identification of the water storage body.
In actual use, the entire model may be evaluated. And (4) evaluating the spectral analysis and the water body classification jointly, and testing the water body identification precision of the whole model. The water body identification precision tau is defined as shown in formula (12).
Figure BDA0001923934930000101
Wherein the content of the first and second substances,
k1+k2+k3＝1 (13)
therefore, the method of the invention realizes the rapid and accurate identification of the reservoir and the small reservoir. The existing target detection method is mainly a target identification method based on region suggestion (region pro-posal). The method based on the regional suggestion needs to use an additional network structure to generate suggestion frames (propusals), the number of the suggestion frames is generally about 300 for each picture, and the calculation cost for generating the suggestion frames is huge, so that the target detection network structure is complex, the running speed is slow, is only about 5fps, and the accuracy is not high, and is about 70%.
The method extracts water area boundary information by using spectral analysis with low calculation cost, frames the water body boundary, determines the position information of the water body in one step, and classifies the water body by using a convolutional neural network. Compared with the traditional method for predicting the water body position by using the suggestion box regression, the method greatly improves the accuracy of water body identification, and enables the accuracy rate to reach more than 80%. Compared with a region generation network RPN + a region-based convolutional neural network R-CNN used in the traditional method, the water body classification network provided by the invention uses a simple convolutional neural network, the network provided by the invention is obviously simplified in structure, the operation speed and the robustness are greatly improved, and the processing speed can reach 23 fps.
The specific embodiment is as follows:
in this embodiment, the region of the middle and upper reaches of the hanjiang river, the red river mouth, is selected as a research area, and the high-definition satellite images of the reservoir and the small reservoir in the research area are shown in fig. 10 and 11.
Firstly, the method provided by the invention is adopted to construct a convolutional neural network, the weight of the network is randomly initialized by adopting normal distribution, the mean value is 0, the variance is 0.1, the initial value of a bias term is 0, the initial value of a learning rate is 0.0001, and the training step number is set as 100K. The learning rate is 0.00002 after 30K of iteration and 0.00001 after 80K of iteration. The network has 5 convolutional layers and a full-connection layer, and the number of feature maps output by each convolutional layer is 64,128,256,512,512 respectively. And (3) expanding the characteristics into 4096-dimensional vectors at a full connection layer, and finally outputting the dimension of the network to be 3 to represent the probability of 3 ground objects such as dams, gates and spillways. And taking the ground object with the highest probability as the result of the classification. The network structure is shown in fig. 7.
And training and testing are carried out by utilizing a data set, the number of pictures contained in the data set is 60000, the proportion of the training set to the testing set is 9:1, namely the testing set contains 54000 pictures, and the testing set contains 6000 pictures. The size of each picture is 224 × 224 pixels, and the pictures are grayscale pictures. The training set and the testing set both contain three types of ground features, namely dams, gates and spillways. As shown in fig. 8, the first row is a gate, the second row is a dam, and the third row is a spillway. The file structure of the data set is shown in fig. 9.
The neural network is trained using the data of the training set, and the number of pictures used for training at a time is 128, i.e., min _ batch equals 128. The dynamic learning rate is adopted, namely when the number of iteration steps is less than 30K, the learning rate is 0.0001, when the number of iteration steps reaches 30K, the learning rate is 0.00002, and when the number of iteration steps reaches 80K, the learning rate is 0.00001. The model is evaluated by using the data of the test set, and the main indexes are recall ratio, precision ratio, F1 and mAP. The mAP of the model is more than or equal to 0.7.
Evaluating the accuracy of the trained network by using the test set
Figure BDA0001923934930000111
Then, the obtained evaluation standard precision recall, precision, F1 and mAP are respectively: 0.9,0.82,0.86,0.70.
According to the invention, a threshold-based multi-band spectral relation method is selected to analyze the high-definition satellite remote sensing image of the research area, and the obtained spectral analysis results are shown in fig. 13 and 14, wherein fig. 13 shows a reservoir, and fig. 14 shows a small reservoir.
On the basis of extracting water bodies by using spectral analysis and classifying land features by using a training network, the input of unknown classes is identified, and information such as the class position of the water body is output. And identifying the dam, the gate and the spillway by using the convolutional neural network, and further determining the category of the impounded body by identifying the dam, the gate and the spillway.
And fusing the water body position information obtained by the spectral analysis and the water body category information obtained by the convolutional neural network to determine the category and the position of the water storage body and realize the identification of the water storage body.
By adopting the technical scheme disclosed by the invention, the following beneficial effects are obtained: the invention provides a reservoir or small reservoir identification method based on spectral analysis and a deep convolutional neural network, which comprises the steps of firstly processing a high-definition remote sensing image by utilizing the spectral analysis, extracting water area boundary information, mapping the water area boundary information onto an original image, then dividing an original image into a plurality of picture blocks containing a water storage body by utilizing the water area boundary information on the original image, taking the picture blocks as the input of a constructed convolutional neural network, identifying and classifying a dam, a gate and a spillway by utilizing the convolutional neural network, and finally identifying and classifying the water bodies (the reservoir and the small reservoir) by utilizing the dam, the gate and the spillway. Therefore, the method provided by the invention has the advantages that the accuracy rate reaches more than 80%, the accuracy of water body identification is greatly improved, the structure of the convolutional neural network is obviously simplified, the running speed and the robustness are greatly improved, the processing speed can reach 23fps, and the rapid and accurate identification of reservoirs and small reservoirs is realized.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and improvements can be made without departing from the principle of the present invention, and such modifications and improvements should also be considered within the scope of the present invention.

Claims (4)

1. A reservoir or small reservoir identification method based on spectral analysis and a deep convolutional neural network is characterized by comprising the following steps:
s1, constructing a convolutional neural network, wherein the network sequentially comprises an input layer, five convolutional layers, a fusion layer, a full-connection layer and an output layer, and the fusion layer is used for fusing the characteristics of the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer;
s2, analyzing the high-resolution satellite remote sensing image by adopting a multi-band spectral relation method based on a threshold value, extracting water area boundary information, framing a water body, and obtaining an image after spectral analysis processing;
s3, mapping the coordinates of the original image and the image after spectral analysis to the same coordinate system, mapping the water area boundary information to the original image under the same coordinate system, and realizing framing of the water body on the original image;
s4, dividing the original picture into a plurality of picture blocks containing water storage bodies by using the water area boundary information on the original image, and taking the picture blocks as the input of a convolutional neural network;
s5, extracting the features of the input picture block containing the water storage body by using the convolutional neural network, and extracting the feature in the input picture block: identifying a dam, a gate or a spillway;
s6, according to the output result of the convolution neural network, namely whether there is a dam, a gate and a spillway, further classifying the input picture blocks, determining the specific type of the water body, namely a reservoir or a small reservoir, and combining the boundary information of the water body to realize the identification of the water body;
the step of S1 further includes: training the convolutional neural network by using a training set, and evaluating the accuracy of the trained convolutional neural network by using a test set;
the evaluation criteria are recall, precision, F1 and mAP, which are calculated according to the following formulas:
Figure FDA0002719557660000011
Figure FDA0002719557660000021
Figure FDA0002719557660000022
Figure FDA0002719557660000023
in the formulas (1) and (2), True positive represents a positive sample predicted to be positive by the model, namely the True accuracy; false negative represents a positive sample predicted to be negative by the model, namely a False negative rate; false positive represents a negative sample predicted to be positive by the model, namely a False alarm rate;
in the formula (4), N is the number of all pictures in the test set, M is the total number of target categories, p (k) represents the value of precision when k pictures can be identified, and Δ r (k) represents the change situation of recall when the number of identified pictures changes from k-1 to k
The method also comprises the step of evaluating the precision of the spectral analysis, wherein the specific evaluation indexes are the proportion rho of the water quantity detected by the spectral analysis to the actual water quantity and the precision determined by the water area boundary frame
Figure FDA0002719557660000026
The proportion calculation formula of the detected water quantity in the actual water quantity is as follows:
wherein the actual water quantity is delta, and the detected water quantity is epsilon;
Figure FDA0002719557660000024
accuracy of water area bounding box determination
Figure FDA0002719557660000025
The method is an index for measuring the extraction accuracy of the boundary information of the water area, and (x, y) is set to represent the coordinate of the upper left corner of the boundary frame alpha of the actual water area, and h and w respectively represent the width and the height of the alpha; (x)1,y1) Coordinates, h, representing the upper left corner of the spectral analysis water bounding box beta1，w1The transformation of β to α can be expressed by equations (7) (8) (9) (10):
Figure FDA0002719557660000031
Figure FDA0002719557660000032
Figure FDA0002719557660000033
Figure FDA0002719557660000034
accuracy of boundary framing in water area
Figure FDA0002719557660000035
Is as defined in formula (11):
Figure FDA0002719557660000036
the method further comprises the step of judging the water body identification precision, wherein the water body identification precision tau is defined as the following formula:
Figure FDA0002719557660000037
wherein the content of the first and second substances,
k1+k2+k3＝1
。
2. the method of claim 1, wherein the training set and the testing set are high-definition image data sets comprising dams, gates and spillways, each image is 224 x 224 pixels in size and is a common gray image, the ratio of the training set to the testing set is 9:1, and the maximum number of training iteration steps is set to 100K.
3. The method for reservoir or small dam identification based on spectral analysis and deep convolutional neural network of claim 2, wherein the convolutional neural network is trained using a training set, specifically: the number of pictures used for training at a time is 128, i.e., min _ batch is 128; the dynamic learning rate is adopted, when the number of iteration steps is less than 30K, the value of the learning rate is 0.0001, when the number of iteration steps reaches 30K, the value of the learning rate is 0.00002, and when the number of iteration steps reaches 80K, the value of the learning rate is 0.00001.
4. The method for identifying a reservoir or a small reservoir based on spectral analysis and deep convolutional neural network as claimed in claim 1, wherein S2 is specifically implemented by analyzing the high-resolution satellite remote sensing image by using the following formula:
(b3+b4)-(b5+b6)>T (5)
in the formula, b3 is the gray value of the wave band 3; b4 is the gray value of band 4; b5 is the gray value of wave band 5; b6 is the gray value of wave band 6; t is a threshold value selected through experiments;
the band is calculated by using the formula, and the water body is obtained when the requirement is met.
CN201811607578.8A 2018-12-27 2018-12-27 Reservoir or small reservoir identification method based on spectral analysis and deep convolutional neural network Active CN109740485B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811607578.8A CN109740485B (en) 2018-12-27 2018-12-27 Reservoir or small reservoir identification method based on spectral analysis and deep convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811607578.8A CN109740485B (en) 2018-12-27 2018-12-27 Reservoir or small reservoir identification method based on spectral analysis and deep convolutional neural network

Publications (2)

Publication Number Publication Date
CN109740485A CN109740485A (en) 2019-05-10
CN109740485B true CN109740485B (en) 2021-02-02

Family

ID=66361373

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811607578.8A Active CN109740485B (en) 2018-12-27 2018-12-27 Reservoir or small reservoir identification method based on spectral analysis and deep convolutional neural network

Country Status (1)

Country Link
CN (1) CN109740485B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110826394A (en) * 2019-07-16 2020-02-21 北京大学 Reservoir identification method and device based on convolutional neural network algorithm
CN112766146B (en) * 2021-01-14 2022-11-29 黄河勘测规划设计研究院有限公司 Multi-source data-based dynamic reservoir monitoring system and method
CN112750138B (en) * 2021-01-14 2022-07-15 黄河勘测规划设计研究院有限公司 Method for identifying space distribution of silt dam of yellow river basin
CN113191242A (en) * 2021-04-25 2021-07-30 西安交通大学 Embedded lightweight driver leg posture estimation method based on OpenPose improvement
CN112990169B (en) * 2021-05-20 2021-08-24 天津美腾科技股份有限公司 Coal-rock interface identification method and coal cutting track determination method and device
CN117315446B (en) * 2023-11-29 2024-02-09 江西省水利科学院(江西省大坝安全管理中心、江西省水资源管理中心) Reservoir spillway abnormity intelligent identification method oriented to complex environment

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108647721A (en) * 2018-05-10 2018-10-12 邢汉发 Ground mulching verification method based on network picture automatic identification and system

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103914692B (en) * 2014-04-21 2017-09-19 山东科技大学 A kind of remote sensing recognition method for colliery peripheral ground water system
CN105279556B (en) * 2015-11-05 2017-11-07 国家卫星海洋应用中心 A kind of Enteromorpha detection method and device
CN106485239A (en) * 2016-11-02 2017-03-08 郑州航空工业管理学院 One kind is using one-class support vector machines detection river mesh calibration method
CN107341517B (en) * 2017-07-07 2020-08-11 哈尔滨工业大学 Multi-scale small object detection method based on deep learning inter-level feature fusion
CN107516317B (en) * 2017-08-18 2021-04-27 上海海洋大学 SAR image sea ice classification method based on deep convolutional neural network
CN107680678B (en) * 2017-10-18 2020-12-01 北京航空航天大学 Thyroid ultrasound image nodule diagnosis system based on multi-scale convolution neural network

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108647721A (en) * 2018-05-10 2018-10-12 邢汉发 Ground mulching verification method based on network picture automatic identification and system

Also Published As

Publication number Publication date
CN109740485A (en) 2019-05-10

Similar Documents

Publication Publication Date Title
CN109740485B (en) Reservoir or small reservoir identification method based on spectral analysis and deep convolutional neural network
Jia et al. Detection and segmentation of overlapped fruits based on optimized mask R-CNN application in apple harvesting robot
EP3614308B1 (en) Joint deep learning for land cover and land use classification
CN111914907B (en) Hyperspectral image classification method based on deep learning space-spectrum combined network
CN111310862B (en) Image enhancement-based deep neural network license plate positioning method in complex environment
CN107316307B (en) Automatic segmentation method of traditional Chinese medicine tongue image based on deep convolutional neural network
CN111797712B (en) Remote sensing image cloud and cloud shadow detection method based on multi-scale feature fusion network
CN109559302A (en) Pipe video defect inspection method based on convolutional neural networks
CN110309781B (en) House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion
CN112861690B (en) Multi-method fused remote sensing image change detection method and system
CN109376591B (en) Ship target detection method for deep learning feature and visual feature combined training
CN111191628B (en) Remote sensing image earthquake damage building identification method based on decision tree and feature optimization
CN110751644B (en) Road surface crack detection method
CN109635726B (en) Landslide identification method based on combination of symmetric deep network and multi-scale pooling
CN110147780B (en) Real-time field robot terrain identification method and system based on hierarchical terrain
CN113239830A (en) Remote sensing image cloud detection method based on full-scale feature fusion
CN112818920B (en) Double-temporal hyperspectral image space spectrum joint change detection method
CN112580661A (en) Multi-scale edge detection method under deep supervision
CN110717921A (en) Full convolution neural network semantic segmentation method of improved coding and decoding structure
CN111507416B (en) Smoking behavior real-time detection method based on deep learning
CN111199245A (en) Rape pest identification method
CN114332644B (en) Large-view-field traffic density acquisition method based on video satellite data
CN117611996A (en) Grape planting area remote sensing image change detection method based on depth feature fusion
CN113469097B (en) Multi-camera real-time detection method for water surface floaters based on SSD network
CN113705441A (en) High-spatial-temporal-resolution surface water body extraction method cooperating with multispectral and SAR images

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant