CN110598600A - Remote sensing image cloud detection method based on UNET neural network - Google Patents
Remote sensing image cloud detection method based on UNET neural network Download PDFInfo
- Publication number
- CN110598600A CN110598600A CN201910814865.4A CN201910814865A CN110598600A CN 110598600 A CN110598600 A CN 110598600A CN 201910814865 A CN201910814865 A CN 201910814865A CN 110598600 A CN110598600 A CN 110598600A
- Authority
- CN
- China
- Prior art keywords
- cloud detection
- remote sensing
- sensing image
- network
- cloud
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a remote sensing image cloud detection method based on a UNET neural network, which comprises the following steps: establishing a cloud detection network with 5 down-sampling layers and 5 up-sampling layers, wherein the front four down-sampling layers of the cloud detection network are connected with a convolution layer and a pooling layer after each layer, and the back four up-sampling layers are connected with an anti-convolution layer after each layer; carrying out cloud labeling, manual rechecking and data enhancement processing on an original remote sensing image set, and dividing the processed remote sensing image set into a training set, an evaluation set and a test set; continuously optimizing the cloud detection network by using the training set and the test set; and carrying out cloud detection on the remote sensing image by utilizing the cloud detection network, and outputting a result picture of the cloud detection. The method solves the problem that the detection result is not ideal due to insufficient cloud feature extraction, improves the detection precision and enhances the universality of the algorithm.
Description
Technical Field
The invention relates to the field of remote sensing image detection, in particular to a remote sensing image cloud detection method based on a UNET neural network.
Background
With the rapid development of satellite remote sensing technology, remote sensing images are widely applied in the fields of environment, agriculture, meteorology and the like. However, satellite remote sensing images of optical imaging are often easily influenced by weather, so that the images are shielded by clouds, and further application and analysis of the images are influenced; images can be shot only when weather conditions permit and no cloud exists, timeliness is affected, and the running cost of the satellite is increased.
In the prior art, various parameters need to be extracted from the picture or the picture needs to be transformed and mapped, and then the threshold is judged, so that the process is very complex, the requirement on image data is high, the fault tolerance is not high, and the popularization is not facilitated. In the prior art, images with different backgrounds are detected with different success rates, certain requirements are placed on the precision of a spectrum and the pretreatment level, otherwise, the error is large, and the adaptability is not strong.
Disclosure of Invention
In view of this, the main object of the present invention is to provide a remote sensing image cloud detection method based on UNET neural network, which solves the problem of unsatisfactory detection result due to insufficient cloud feature extraction, improves the detection accuracy, and enhances the universality of the algorithm.
The invention discloses a remote sensing image cloud detection method based on a UNET neural network, which has the specific technical scheme that:
establishing a cloud detection network with 5 down-sampling layers, 5 up-sampling layers and 4 jump connecting chains, wherein a pooling layer is connected behind each of the first four down-sampling layers of the cloud detection network, and an anti-convolution layer is connected behind each of the last four up-sampling layers;
carrying out cloud labeling, manual rechecking and data enhancement processing on an original remote sensing image set, and dividing the processed remote sensing image set into a training set, an evaluation set and a test set;
training the cloud detection network by using the images in the training set, and continuously updating the parameters of the cloud detection network;
carrying out accuracy test on the trained cloud detection network by using the images in the test set, and carrying out parameter adjustment and retraining on the cloud detection network;
and carrying out cloud detection on the remote sensing image by utilizing the cloud detection network, and outputting a result picture of the cloud detection.
Further, comprising: and a fifth down-sampling layer of the cloud detection network is connected with the first up-sampling layer.
Further, comprising: and the last upsampling layer of the cloud detection network is connected with the full connection layer.
Further, cloud labeling of the original remote sensing image set comprises: and carrying out cloud annotation on the remote sensing image set by utilizing ENVI software.
Further, the data enhancement of the original remote sensing image set comprises: and carrying out rotation translation, elastic deformation and contrast enhancement on the remote sensing image set.
Further, the accuracy test is performed on the trained cloud detection network by using the image in the test set, and the parameter adjustment and retraining of the cloud detection network comprises: and adjusting the training times and the training batch size according to the performance of the loss function in the training.
Further, the dividing the processed remote sensing image set into a training set, an evaluation set and a test set further comprises: and randomly selecting 70% of images from the remote sensing image set as a training set, 20% of images as an evaluation set, and the rest 10% of images as a test set.
Further, comprising: the activation function of the cloud detection network is a ReLU function.
Further, training the cloud detection network by using the images in the training set, and continuously updating parameters of the cloud detection network, including: the parameters of the cloud detection network may be weights of convolution and bias values of convolution.
Further, the accuracy test is performed on the trained cloud detection network by using the image in the test set, and the parameter adjustment and retraining of the cloud detection network are performed, including: the learning rate is set to 0.001.
The method utilizes a deep learning model UNET network to carry out down-sampling on the optical remote sensing image to extract image characteristics, and then combines the up-sampling with some jump connections to generate a target detection image. The problem that detection results are not ideal enough due to insufficient cloud feature extraction is solved, the detection precision is improved, the universality of the algorithm is enhanced, the development of scientific research and economic activities is facilitated, and the method has high intelligence and convenience.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a diagram of a cloud detection network architecture of the present invention;
FIG. 3 is a schematic diagram of a dotted line in a cloud detection network structure according to the present invention;
FIG. 4 is a sample diagram of a training set according to an embodiment of the present invention, in which the left half is an original remote sensing image, and the right half is a corresponding labeled diagram;
FIG. 5 is an original image of a remote sensing image for detection according to an embodiment of the present invention;
FIG. 6 is an annotation diagram after the remote sensing image original image is annotated according to the embodiment of the invention;
fig. 7 is a diagram illustrating a network detection result of an original image of a remote sensing image according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The remote sensing image cloud detection method based on the UNET neural network mainly works by building and training the UNET network, and the network takes a remote sensing image as input and outputs a mask of the remote sensing image cloud. And the construction of the network is realized by utilizing a deep learning platform, and the training of the network comprises the processes of making a data set and training and parameter adjustment. In the embodiment of the invention, the data set is manufactured without matching pictures and labels as in the traditional supervised learning, only a large number of original satellite remote sensing images are collected, the original remote sensing images are subjected to initial cloud labeling by using ENVI software, then manual inspection and modification are carried out, cloud detection mask graphs corresponding to the original remote sensing images one by one are obtained, and the cloud detection mask graphs are used as training sets.
The embodiment of the invention provides a remote sensing image cloud detection method based on a UNET neural network, which specifically comprises the following steps:
firstly, establishing a UNET network model:
the UNET adopts a network structure comprising down-sampling and up-sampling, the UNET network can be simply seen as down-sampling first, deep features are learned through convolution of different degrees, the original image size is reduced through up-sampling, and the up-sampling is realized through deconvolution. The down-sampling is used to gradually present the environment information, and the up-sampling is a process of restoring detail information by combining the down-sampled layer information and the up-sampled input information, and gradually restoring the image precision.
The convolution operation with the same number of layers is adopted in the up-sampling stage and the down-sampling stage of the UNET network, and the down-sampling layer is connected with the up-sampling layer by using a jump connection link structure, so that the characteristics extracted by the down-sampling layer can be directly transmitted to the up-sampling layer, the pixel positioning of the UNET network is more accurate, and the segmentation precision is higher.
As shown in fig. 2, the cloud detection network established in the embodiment of the present invention is based on the UNET neural network, and includes 5 down-sampling layers and 5 up-sampling layers, wherein a pooling layer is connected to each of the first four down-sampling layers, an anti-convolution layer is connected to each of the last four up-sampling layers, and the cloud detection network has four hopping connection chains, the fifth down-sampling layer is connected to the first up-sampling layer, the last up-sampling layer is connected to the full connection layer for output, and the activation function is a ReLU function.
The down-sampling layer is used for extracting a high-dimensional characteristic diagram from a matrix corresponding to the remote sensing image through a large number of convolution kernels, so that a plurality of characteristic diagrams can be extracted, each characteristic diagram is local perception extracted from the image, and interested parts in the image can be extracted by integrating the characteristic diagrams. The embodiment of the invention adopts convolution kernels with the size of 3 x 3 and the step length of 1, changes the image in a 3 x 3 window in the original remote sensing image into one pixel, performs convolution twice on each convolution layer, doubles the number of the convolution kernels of each layer after each downsampling, and makes the number of the convolution kernels of the next downsampling layer 2 times that of the convolution kernels of the previous downsampling layer, thereby effectively extracting the characteristic information of the image and realizing the multi-scale characteristic identification of the image characteristic.
In the multilayer neural network, the excitation function is used for carrying out nonlinear mapping on the feature vector diagram obtained by convolution, and after multilayer convolution, the value of the feature vector diagram does not change greatly during training, so that gradient disappears, and further a convolution network model cannot be trained. After each convolution, the ReLU excitation function is used, and nonlinear mapping can be carried out on the linear calculation result of the convolution layer, so that the change of the characteristic diagram before and after training is not too small, and further the convolution network model can be used for training.
For the first four convolutional layers, a pooling layer is connected behind each convolutional layer, and the functions of feature dimension reduction, data and parameter quantity compression, training overfitting reduction and improvement of the fault tolerance of the cloud detection network are achieved. The embodiment of the invention uses pooling nuclei with size 2 x 2 and step size 2, and the length and width of the pooled feature map are reduced by half.
The function of the upsampling layer is to restore the picture from the feature vector diagram of the picture. The feature map of each upper sampling layer and the feature map of the corresponding lower sampling layer are subjected to jump connection, namely splicing shown in the attached drawing 2 and then convolution are carried out, the upper sampling layer is also a convolution kernel with the size of 3 x 3 and the step length of 2, so that the aim of enabling the picture to refer to features with lower dimensionality in the reduction process is that the reduced picture is closer to an original picture. The number of convolution kernels of each up-sampling layer corresponds to the number of convolution kernels of the down-sampling layer on the same layer.
The effect of the deconvolution layer can be regarded as the inverse effect of the pooling layer, and the feature map is extended by interpolation using a deconvolution kernel with a size of 2 x 2 and a step size of 2.
Full connection layer: and the last layer of the upper sampling layer is connected with two full-connection layers, the full-connection layers use convolution kernels with the size of 1 x 1 to respectively obtain the foreground and the background of the segmentation graph, and then the segmentation graph and the background are combined to output a result mask.
As shown in fig. 2, the cloud detection network according to the embodiment of the present application has 5 down-sampling layers and 5 up-sampling layers, a convolutional layer and a pooling layer are connected after each of the first 4 down-sampling layers, the 5 th down-sampling layer is connected with the 1 st up-sampling layer, an anti-convolutional layer is connected after each of the last 4 up-sampling layers, and a jump connection link is provided between the up-sampling layer and the down-sampling layer of the corresponding layer in the first 4 layers. Inputting a remote sensing image with any size into a cloud detection network, extracting features by using a down-sampling layer, extracting a high-dimensional feature map from a matrix of the remote sensing image by using a convolution kernel with the size of 3 x 3 and the step length of 1, performing convolution operation for each layer for 2 times, and performing nonlinear mapping on a feature vector map obtained by convolution by using an excitation function after each convolution. The first downsampling layer uses 32 convolution kernels, after 2 times of convolution operation, the number of data and parameters to be processed in the next downsampling layer is reduced by using the pooling layer, the pooling layer of the embodiment of the invention uses a filter with the size of 2 x 2 and the step length of 2, the length and the width of the feature map after pooling are 1/2 of the original feature map, namely, the data amount after pooling is reduced to 1/4 before pooling, so that network overfitting can be prevented. The down-sampling layer is used as a feature extraction part, one scale is needed after passing through one pooling layer, the embodiment of the application comprises an original image scale, the original image scale is 5 scales in total, and the image feature multi-scale feature identification by a network is realized by 4 pooling layers.
The number of convolution kernels of the next down-sampling layer is 2 times of the number of convolution kernels of the previous down-sampling layer, as shown in fig. 2, the number of convolution kernels of the second down-sampling layer is 64, the number of convolution kernels of the third down-sampling layer is 128, the number of convolution kernels of the fourth down-sampling layer is 256, the number of convolution kernels of the last down-sampling layer is 512, and the last down-sampling layer is connected with the first up-sampling layer. And the up-sampling part splices the feature map of the down-sampling layer of the corresponding level every time up-sampling is carried out, then splices the feature map of the last up-sampling layer with the feature map of the down-sampling layer of the corresponding level, and carries out convolution operation twice by adopting a convolution kernel with the size of 3 x 3 and the step length of 2. After the two convolution operations of each upper sampling layer are finished, performing deconvolution operation, and expanding the characteristic graph by adopting an interpolation method. And connecting the last upper sampling layer with two fully-connected layers, respectively obtaining the foreground and the background of the segmentation image by using a convolution kernel with the size of 1 x 1 in the fully-connected layers, and then combining and outputting the results to obtain a mask image of the original remote sensing image.
As shown in fig. 3, the upsampling part fuses the output of the feature extraction part, which actually fuses the multi-scale features together, taking the second upsampling layer as an example, the features of the second upsampling layer come from the features of the output of the 4 th downsampling layer, i.e. the dashed rectangle part on the right of the dashed arrow in fig. 3, and also come from the output of the first upsampling layer, and such splicing is performed throughout the entire cloud detection network, and there are four times of splicing processes in the network of the embodiment of the present application.
The feature map obtained by each convolution layer of the UNET network in the embodiment of the invention is connected to the corresponding upsampling layer, so that the feature map of each layer is effectively used in subsequent calculation, supervision and loss calculation in a high-level feature map are avoided, and the features in a low-level feature map are combined, so that the finally obtained feature map not only contains high-level features, but also protects a plurality of low-level features, the fusion of the features at different levels is realized, and the result accuracy of the model is improved.
Secondly, the manufacturing process of the data set:
in the embodiment of the application, the Landsat satellite pictures are used as data sets, and the backgrounds are roughly divided into five picture sets of snowfield, grassland, desert, ocean and town;
according to the method and the device, multispectral image sets and sensor data of corresponding images in the Landsat data sets are utilized, Cloud layer identification tools in ENVI5.4 software are used, the Size (Kernel Size) of outward expansion of Cloud areas and Cloud-free land and water area possibility Threshold (Cloud Prohability Threshold) parameters are adjusted, the Cloud in the images can be preliminarily identified, and the possibility of false judgment and missed judgment still exists. Wherein the larger the likelihood threshold for cloud-cloudless land and water areas, the smaller the cloud area that is likely to be detected.
The method comprises the steps of carrying out manual rechecking on a picture after preliminary automatic processing, complementing a mask picture with cloud misjudgment and missed judgment by using a picture editing tool to serve as a training label, wherein each original remote sensing image is provided with a cloud detection mask image which is manually marked and corresponds to the original remote sensing image, the original remote sensing image and the cloud detection mask image corresponding to the original remote sensing image form a training data set, such as a training set sample shown in figure 3, the original remote sensing image is arranged on the left side, and the marked image corresponding to the original remote sensing image is arranged on the right side.
In the embodiment of the application, 460 images obtained after manual labeling are rotated, deformed, and contrast enhancement are performed on each image by using an openCV (constant value library) of python, the original image and the mask image can be rotated at random angles, for example, rotated by 90 degrees or rotated by 180 degrees, the images are deformed by using a getAffieTransform function, and the contrast of the images is enhanced by enhancing the maximum value and the minimum value of the contrast. And (4) carrying out binarization processing on part of pictures, wherein the binarization effect of some pictures has no significance, such as the connection of clouds and water, so that the binarization processing is not carried out on all the pictures.
A certain number of picture sets are obtained through data expansion and enhancement, namely, rotation and translation, elastic deformation, gray value change and the like are performed to a certain degree, diversity of the data sets is enriched, and robustness of a network model is enhanced.
If a new data set is added, the cloud detection result graph can be obtained by performing rough detection and manual verification by using ENVi software.
Thirdly, the process of network training by using the data set is as follows:
and (3) training the remote sensing image data set by using a Dropout method, repeatedly training the constructed network, and indicating that the constructed UNET network model meets the requirements after the training times reach a preset threshold value or the testing accuracy reaches a target value.
The loss function of the UNET network model in the embodiment of the present application is:
wherein E is a cross entropy calculation function, x is a pixel point of the input picture, pl(x)To approximate a maximum function, wc(x) As class frequency weight, d1(x) Representing the distance of a pixel point x to the boundary of the cloud zone nearest to it, d2(x) Representing the distance of a pixel point x to the cloud boundary second closest thereto, parameter w0And the value of the standard deviation sigma is adjustable.
Calculating error loss by using loss function to estimate difference between two vectors, wherein the loss function is approximate to maximum function pl(x)The objects of (1) are the artificial annotation graph and the image output by the cloud detection network. In the embodiment of the application, the difference between the mask generated by the cloud detection network and the real annotation image is evaluated by using the loss function, and the difference is fed back to the cloud detection network for parameter updating in the training process. In order to make the result of the loss function closer to the requirement of practical application, the cross entropy function is weighted to compensate different frequencies of each type of pixels in the training data set.
When the cloud detection network is trained by using the training data set, an original remote sensing picture is input into the current network every time, the output picture of the current network is obtained through forward transmission calculation, the error between the output picture and the corresponding mask picture which is marked manually is calculated by using a loss function, the error is reversely propagated into the network by using a chain rule, and parameters in the network, such as convolution weight, convolution bias and the like, are updated by using an Adam optimizer in the process of reverse propagation in one round to finish one-time learning. In the embodiment of the invention, 1050 original pictures are used as training samples, the size of each picture is 572 x 572, 1050 pictures are trained once during training to be used as a period, and the embodiment of the invention trains 50 periods in total.
Fourthly, adjusting parameters of the network model and retraining:
and inputting the images in the test data set and the evaluation data set into the cloud detection network, testing the accuracy, training the cloud detection network, adjusting the values of parameters such as the weight of convolution, the bias of convolution and the like, and continuously repeating the training process until the expected effect is achieved.
In deep learning, the weight initialization method of the neural network has a crucial influence on the convergence speed and the performance of the model, and the neural network continuously and iteratively updates the weight parameters so as to achieve better performance. Parameter learning in the training process of the neural network is optimized based on a gradient descent method. The gradient descent method requires an initial value to be assigned to each parameter at the start of training. In practical application, the parameters obey a more effective initialization mode of Gaussian distribution or uniform distribution. In the deep neural network, as the number of layers increases, gradient disappearance or gradient explosion easily occurs in the process of gradient descent. The weight can be initialized in the form of all 0, all 1 or a fixed value, and in the embodiment of the invention, the weight is initialized by using the fixed value, and the adopted fixed value is 0.1.
According to the performance of the training loss function, the training times and the training batch size can be adjusted, and the training speed is considered during adjustment. The learning rate is adjusted, the network can enter local optimization too early due to an excessive rate, the effect is not ideal, and the learning rate of the embodiment of the invention is finally set to be 0.001.
The invention adopts the deep learning network for detection, saves the design and realization of complex algorithm and can more conveniently and directly achieve the aim. Meanwhile, when the existing artificial intelligence method is used for processing the problems, the network structure is usually too simple or limited by the structure, the higher precision cannot be achieved by increasing the number of layers, and the condition of missed judgment and erroneous judgment is easy to occur. Compared with the traditional multilayer convolution full-connection network, the UNET up-sampling still has a large number of channels and down-sampling connection, so that the network transmits the context information to a higher layer resolution, as a result, the expansion path and the contraction path are symmetrical to form a U-shaped shape, the details of the bottom layer are effectively reserved, and the detection with higher precision is carried out.
The remote sensing image cloud detection method based on the UNET neural network is a method for directly inputting a picture to outputting a picture mask, and the up-sampling of the UNET neural network still has a large number of channels connected with the down-sampling, so that the network transmits context information to a higher layer of resolution, an expansion path and a contraction path are symmetrical to form a U-shaped shape, the details of a bottom layer are effectively reserved, and the detection with higher precision is carried out.
On the basis of combining the use of the deep learning convolutional neural network, the cloud detection is carried out on the remote sensing picture, then the time-sharing shooting is carried out on the same region, the region with the cloud and the region without the cloud are replaced and integrated into the image without the cloud, and the key points of the method comprise that: the remote sensing image is directly processed without being used for preprocessing the previous picture or pre-extracting parameters; the mask picture is directly output without later generation.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A remote sensing image cloud detection method based on a UNET neural network is characterized by comprising the following steps:
establishing a cloud detection network with 5 down-sampling layers and 5 up-sampling layers, wherein the front four down-sampling layers of the cloud detection network are connected with a convolution layer and a pooling layer after each layer, and the back four up-sampling layers are connected with an anti-convolution layer after each layer;
carrying out cloud labeling, manual rechecking and data enhancement processing on an original remote sensing image set, and dividing the processed remote sensing image set into a training set, an evaluation set and a test set;
training the cloud detection network by using the images in the training set, and continuously updating the parameters of the cloud detection network;
carrying out accuracy test on the trained cloud detection network by using the images in the test set, and carrying out parameter adjustment and retraining on the cloud detection network;
and carrying out cloud detection on the remote sensing image by utilizing the cloud detection network, and outputting a mask image of the cloud detection.
2. The UNET neural network-based remote sensing image cloud detection method according to claim 1, comprising the following steps: and a fifth down-sampling layer of the cloud detection network is connected with the first up-sampling layer.
3. The UNET neural network-based remote sensing image cloud detection method according to claim 1, comprising the following steps: and the last upsampling layer of the cloud detection network is connected with the output of the full connection layer.
4. The remote sensing image cloud detection method based on the UNET neural network as claimed in claim 1, wherein cloud labeling of the original remote sensing image set comprises: and carrying out cloud annotation on the remote sensing image set by utilizing ENVI software.
5. The UNET neural network-based remote sensing image cloud detection method according to claim 1, wherein the data enhancement of the original remote sensing image set comprises: and carrying out rotation translation, elastic deformation and contrast enhancement on the remote sensing image set.
6. The UNET neural network-based remote sensing image cloud detection method according to claim 1, wherein the accuracy test is performed on the trained cloud detection network by using the image in the test set, and the parameter adjustment and retraining of the cloud detection network comprises: and adjusting the training times and the training batch size according to the performance of the loss function in the training.
7. The UNET neural network-based remote sensing image cloud detection method according to claim 1, wherein the dividing the processed remote sensing image set into a training set, an evaluation set, and a test set further comprises: and randomly selecting 70% of images from the remote sensing image set as a training set, 20% of images as an evaluation set, and the rest 10% of images as a test set.
8. The UNET neural network-based remote sensing image cloud detection method according to claim 1, comprising: the activation function of the cloud detection network is a ReLU function.
9. The UNET neural network-based remote sensing image cloud detection method according to claim 1, wherein the cloud detection network is trained by using the images in the training set, and parameters of the cloud detection network are continuously updated, and the method includes: the parameters of the cloud detection network may be weights of convolution and bias values of convolution.
10. The UNET neural network-based remote sensing image cloud detection method according to claim 1, wherein the accuracy test is performed on the trained cloud detection network by using the image in the test set, and parameter adjustment and retraining of the cloud detection network are performed, and the method includes: the learning rate is set to 0.001.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910814865.4A CN110598600A (en) | 2019-08-27 | 2019-08-27 | Remote sensing image cloud detection method based on UNET neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910814865.4A CN110598600A (en) | 2019-08-27 | 2019-08-27 | Remote sensing image cloud detection method based on UNET neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110598600A true CN110598600A (en) | 2019-12-20 |
Family
ID=68856789
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910814865.4A Pending CN110598600A (en) | 2019-08-27 | 2019-08-27 | Remote sensing image cloud detection method based on UNET neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110598600A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111428726A (en) * | 2020-06-10 | 2020-07-17 | 中山大学 | Panorama segmentation method, system, equipment and storage medium based on graph neural network |
| CN111553289A (en) * | 2020-04-29 | 2020-08-18 | 中国科学院空天信息创新研究院 | Remote sensing image cloud detection method and system |
| CN111611932A (en) * | 2020-05-22 | 2020-09-01 | 哈尔滨工业大学（深圳） | Remote sensing image cloud detection method based on full convolution network, terminal and storage medium |
| CN111915592A (en) * | 2020-08-04 | 2020-11-10 | 西安电子科技大学 | Remote sensing image cloud detection method based on deep learning |
| CN111951400A (en) * | 2020-08-06 | 2020-11-17 | 天津大学 | Narrow-beam X-ray excitation luminescence tomography method based on U-net network |
| CN112101309A (en) * | 2020-11-12 | 2020-12-18 | 北京道达天际科技有限公司 | Ground object target identification method and device based on deep learning segmentation network |
| CN112348744A (en) * | 2020-11-24 | 2021-02-09 | 电子科技大学 | Data enhancement method based on thumbnail |
| CN112729562A (en) * | 2021-01-26 | 2021-04-30 | 河南工业大学 | Sea ice distribution detection method based on improved U-shaped convolutional neural network |
| CN113034498A (en) * | 2021-04-28 | 2021-06-25 | 江苏欧密格光电科技股份有限公司 | LED lamp bead defect detection and assessment method and device, computer equipment and medium |
| CN113392915A (en) * | 2021-06-23 | 2021-09-14 | 宁波聚华光学科技有限公司 | Industrial part defect detection method based on deep learning |
| CN113421258A (en) * | 2021-07-22 | 2021-09-21 | 湖南星图空间信息技术有限公司 | Automatic cloud detection method based on high-resolution remote sensing image |
| CN113569926A (en) * | 2021-07-13 | 2021-10-29 | 中国资源卫星应用中心 | Training method and device for cloud segmentation model with high mobility |
| CN113792653A (en) * | 2021-09-13 | 2021-12-14 | 山东交通学院 | Method, system, equipment and storage medium for cloud detection of remote sensing image |
| CN114494821A (en) * | 2021-12-16 | 2022-05-13 | 广西壮族自治区自然资源遥感院 | Remote sensing image cloud detection method based on feature multi-scale perception and self-adaptive aggregation |
| CN116071622A (en) * | 2023-04-06 | 2023-05-05 | 广州思德医疗科技有限公司 | Stomach image recognition model construction method and system based on deep learning |
| CN117036984A (en) * | 2023-10-09 | 2023-11-10 | 武汉大学 | Cascade U-shaped network cloud detection method and system integrating attention mechanisms |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109255294A (en) * | 2018-08-02 | 2019-01-22 | 中国地质大学（北京） | A kind of remote sensing image clouds recognition methods based on deep learning |
| CN110119728A (en) * | 2019-05-23 | 2019-08-13 | 哈尔滨工业大学 | Remote sensing images cloud detection method of optic based on Multiscale Fusion semantic segmentation network |
-
2019
- 2019-08-27 CN CN201910814865.4A patent/CN110598600A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109255294A (en) * | 2018-08-02 | 2019-01-22 | 中国地质大学（北京） | A kind of remote sensing image clouds recognition methods based on deep learning |
| CN110119728A (en) * | 2019-05-23 | 2019-08-13 | 哈尔滨工业大学 | Remote sensing images cloud detection method of optic based on Multiscale Fusion semantic segmentation network |
Non-Patent Citations (1)
| Title |
|---|
| JACOB HØXBROEJEPPESEN，ET.AL: "A cloud detection algorithm for satellite imagery based on deep learning", 《REMOTE SENSING OF ENVIRONMENT》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111553289A (en) * | 2020-04-29 | 2020-08-18 | 中国科学院空天信息创新研究院 | Remote sensing image cloud detection method and system |
| CN111611932A (en) * | 2020-05-22 | 2020-09-01 | 哈尔滨工业大学（深圳） | Remote sensing image cloud detection method based on full convolution network, terminal and storage medium |
| CN111428726B (en) * | 2020-06-10 | 2020-09-11 | 中山大学 | Panorama segmentation method, system, equipment and storage medium based on graph neural network |
| CN111428726A (en) * | 2020-06-10 | 2020-07-17 | 中山大学 | Panorama segmentation method, system, equipment and storage medium based on graph neural network |
| CN111915592B (en) * | 2020-08-04 | 2023-08-22 | 西安电子科技大学 | Remote sensing image cloud detection method based on deep learning |
| CN111915592A (en) * | 2020-08-04 | 2020-11-10 | 西安电子科技大学 | Remote sensing image cloud detection method based on deep learning |
| CN111951400A (en) * | 2020-08-06 | 2020-11-17 | 天津大学 | Narrow-beam X-ray excitation luminescence tomography method based on U-net network |
| CN112101309A (en) * | 2020-11-12 | 2020-12-18 | 北京道达天际科技有限公司 | Ground object target identification method and device based on deep learning segmentation network |
| CN112348744A (en) * | 2020-11-24 | 2021-02-09 | 电子科技大学 | Data enhancement method based on thumbnail |
| CN112729562B (en) * | 2021-01-26 | 2022-08-26 | 河南工业大学 | Sea ice distribution detection method based on improved U-shaped convolutional neural network |
| CN112729562A (en) * | 2021-01-26 | 2021-04-30 | 河南工业大学 | Sea ice distribution detection method based on improved U-shaped convolutional neural network |
| CN113034498A (en) * | 2021-04-28 | 2021-06-25 | 江苏欧密格光电科技股份有限公司 | LED lamp bead defect detection and assessment method and device, computer equipment and medium |
| CN113034498B (en) * | 2021-04-28 | 2023-11-28 | 江苏欧密格光电科技股份有限公司 | LED lamp bead defect detection and assessment method, device, computer equipment and medium |
| CN113392915A (en) * | 2021-06-23 | 2021-09-14 | 宁波聚华光学科技有限公司 | Industrial part defect detection method based on deep learning |
| CN113569926A (en) * | 2021-07-13 | 2021-10-29 | 中国资源卫星应用中心 | Training method and device for cloud segmentation model with high mobility |
| CN113421258A (en) * | 2021-07-22 | 2021-09-21 | 湖南星图空间信息技术有限公司 | Automatic cloud detection method based on high-resolution remote sensing image |
| CN113792653A (en) * | 2021-09-13 | 2021-12-14 | 山东交通学院 | Method, system, equipment and storage medium for cloud detection of remote sensing image |
| CN113792653B (en) * | 2021-09-13 | 2023-10-20 | 山东交通学院 | Method, system, equipment and storage medium for cloud detection of remote sensing image |
| CN114494821B (en) * | 2021-12-16 | 2022-11-18 | 广西壮族自治区自然资源遥感院 | Remote sensing image cloud detection method based on feature multi-scale perception and self-adaptive aggregation |
| CN114494821A (en) * | 2021-12-16 | 2022-05-13 | 广西壮族自治区自然资源遥感院 | Remote sensing image cloud detection method based on feature multi-scale perception and self-adaptive aggregation |
| CN116071622A (en) * | 2023-04-06 | 2023-05-05 | 广州思德医疗科技有限公司 | Stomach image recognition model construction method and system based on deep learning |
| CN116071622B (en) * | 2023-04-06 | 2024-01-12 | 广州思德医疗科技有限公司 | Stomach image recognition model construction method and system based on deep learning |
| CN117036984A (en) * | 2023-10-09 | 2023-11-10 | 武汉大学 | Cascade U-shaped network cloud detection method and system integrating attention mechanisms |
| CN117036984B (en) * | 2023-10-09 | 2024-01-09 | 武汉大学 | Cascade U-shaped network cloud detection method and system integrating attention mechanisms |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110598600A (en) | Remote sensing image cloud detection method based on UNET neural network | |
| CN109584248B (en) | Infrared target instance segmentation method based on feature fusion and dense connection network | |
| CN110188685B (en) | Target counting method and system based on double-attention multi-scale cascade network | |
| Guo et al. | CDnetV2: CNN-based cloud detection for remote sensing imagery with cloud-snow coexistence | |
| CN112287978B (en) | Hyperspectral remote sensing image classification method based on self-attention context network | |
| CN110119728B (en) | Remote sensing image cloud detection method based on multi-scale fusion semantic segmentation network | |
| CN111738124B (en) | Remote sensing image cloud detection method based on Gabor transformation and attention | |
| CN109886066B (en) | Rapid target detection method based on multi-scale and multi-layer feature fusion | |
| CN110659591B (en) | SAR image change detection method based on twin network | |
| CN111079683A (en) | Remote sensing image cloud and snow detection method based on convolutional neural network | |
| CN112288011B (en) | Image matching method based on self-attention deep neural network | |
| CN112541904A (en) | Unsupervised remote sensing image change detection method, storage medium and computing device | |
| CN109741340B (en) | Ice cover radar image ice layer refined segmentation method based on FCN-ASPP network | |
| CN109635726B (en) | Landslide identification method based on combination of symmetric deep network and multi-scale pooling | |
| CN115713533B (en) | Power equipment surface defect detection method and device based on machine vision | |
| CN112837320B (en) | Remote sensing image semantic segmentation method based on parallel hole convolution | |
| CN114998566A (en) | Interpretable multi-scale infrared small and weak target detection network design method | |
| CN110991430A (en) | Ground feature identification and coverage rate calculation method and system based on remote sensing image | |
| CN115049945B (en) | Unmanned aerial vehicle image-based wheat lodging area extraction method and device | |
| CN114419406A (en) | Image change detection method, training method, device and computer equipment | |
| CN115512222A (en) | Method for evaluating damage of ground objects in disaster scene of offline training and online learning | |
| CN115861756A (en) | Earth background small target identification method based on cascade combination network | |
| CN110472632B (en) | Character segmentation method and device based on character features and computer storage medium | |
| CN113807356B (en) | End-to-end low-visibility image semantic segmentation method | |
| CN114092803A (en) | Cloud detection method and device based on remote sensing image, electronic device and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191220 |
|
| RJ01 | Rejection of invention patent application after publication |