CN112257496A - Deep learning-based power transmission channel surrounding environment classification method and system - Google Patents
Deep learning-based power transmission channel surrounding environment classification method and system Download PDFInfo
- Publication number
- CN112257496A CN112257496A CN202010960023.2A CN202010960023A CN112257496A CN 112257496 A CN112257496 A CN 112257496A CN 202010960023 A CN202010960023 A CN 202010960023A CN 112257496 A CN112257496 A CN 112257496A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing image
- deep learning
- power transmission
- transmission channel
- 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
- 230000005540 biological transmission Effects 0.000 title claims abstract description 93
- 238000000034 method Methods 0.000 title claims abstract description 80
- 238000013135 deep learning Methods 0.000 title claims abstract description 27
- 238000013136 deep learning model Methods 0.000 claims abstract description 101
- 238000012549 training Methods 0.000 claims abstract description 78
- 238000011176 pooling Methods 0.000 claims description 29
- 238000012360 testing method Methods 0.000 claims description 27
- 238000010586 diagram Methods 0.000 claims description 23
- 230000006870 function Effects 0.000 claims description 20
- 230000011218 segmentation Effects 0.000 claims description 20
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000007637 random forest analysis Methods 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 7
- 230000003993 interaction Effects 0.000 claims description 7
- 230000003213 activating effect Effects 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 description 20
- 238000012545 processing Methods 0.000 description 9
- 238000013527 convolutional neural network Methods 0.000 description 8
- 230000003595 spectral effect Effects 0.000 description 8
- 238000004590 computer program Methods 0.000 description 7
- 238000012544 monitoring process Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 238000009795 derivation Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 235000002566 Capsicum Nutrition 0.000 description 1
- 239000006002 Pepper Substances 0.000 description 1
- 235000016761 Piper aduncum Nutrition 0.000 description 1
- 235000017804 Piper guineense Nutrition 0.000 description 1
- 244000203593 Piper nigrum Species 0.000 description 1
- 235000008184 Piper nigrum Nutrition 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 150000003839 salts Chemical class 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
Abstract
The invention provides a method and a system for classifying surrounding environments of a power transmission channel based on deep learning, wherein the method and the system comprise the following steps: acquiring a high-resolution remote sensing image of the surrounding environment of a power transmission channel to be classified; inputting the high-resolution remote sensing image into a pre-trained deep learning model to generate a class probability map of the surrounding environment of the power transmission channel; classifying the power transmission channel surrounding environment based on the class probability map; the deep learning model is obtained by training by using an object-oriented classification method based on the historical remote sensing images and the probability of each ground object type in the historical remote sensing images. The classification of the surrounding environment of the power transmission channel is realized by utilizing the deep learning model and the object-oriented method, the high-resolution remote sensing image can be further utilized to classify the ground surface covering objects around the power transmission channel, and meanwhile, basic data support and auxiliary decision basis are provided for the safe operation of power facilities, emergency rescue planning and the like.
Description
Technical Field
The invention relates to the field of deep learning, in particular to a method and a system for classifying surrounding environments of a power transmission channel based on deep learning.
Background
And monitoring the surrounding environment of the power transmission channel, namely monitoring the ground surface coverage type around the power transmission channel. At present, the ground surface coverage type monitoring around the power transmission channel is mainly realized based on an unmanned aerial vehicle, a ground terminal and an online monitoring device, and a large amount of manpower, material resources and financial resources are consumed; meanwhile, the method has great defects in coverage, timeliness, reliability and the like. Satellite remote sensing observation has incomparable huge advantages: the method has the advantages of wide coverage area, high information acquisition speed, short updating period, less limitation of acquired information by conditions, large amount of acquired information and various sensor types, and is particularly suitable for monitoring the surrounding environment of the long-distance power transmission channel.
At present, the monitoring of ground objects around a power transmission channel by using a satellite remote sensing technology is relatively rare, and the traditional pixel classification and object-oriented classification are adopted. The traditional pixel-based classification method has poor classification precision and serious 'salt and pepper' phenomenon of a classification result, and is not suitable for high-resolution images. Although the object-oriented classification method extracts the characteristics of the object such as spectrum, shape and the like, the description of the high-level semantic characteristics such as texture, environment and the like which are the most important factors for recognizing the ground features is not comprehensive enough, and the information quantity is not enough to support complete ground feature classification and recognition. In view of this, the present invention provides a method for "deep learning" to grasp the high-level semantic characteristics of different objects, such as textures and environments, and form a deep learning model to classify the objects, so as to improve the classification accuracy.
Disclosure of Invention
Aiming at the problems that the classification precision is poor based on a pixel classification method and the monitoring efficiency of the surrounding environment of a power transmission channel is low because the classification cannot be completely carried out due to incomplete identified information quantity in the object-oriented classification method in the prior art, the invention provides a method and a system for classifying the surrounding environment of the power transmission channel based on deep learning, which comprises the following steps:
acquiring a high-resolution remote sensing image of the surrounding environment of a power transmission channel to be classified;
inputting the high-resolution remote sensing image into a pre-trained deep learning model to generate a class probability map of the surrounding environment of the power transmission channel;
classifying the power transmission channel surrounding environment based on the class probability map;
the deep learning model is obtained by training by using an object-oriented classification method based on the historical remote sensing images and the probability of each ground object type in the historical remote sensing images.
Preferably, the training of the deep learning model comprises:
acquiring a historical remote sensing image of the surrounding environment of a power transmission channel;
carrying out multi-scale segmentation on the historical remote sensing image of the surrounding environment of the power transmission channel by using a set scale to obtain a multi-region remote sensing image;
selecting a regional remote sensing image containing all ground feature categories from the multi-region remote sensing image as a sample region;
carrying out full-element classification on the sample area by using an object-oriented classification method to obtain the probability of each ground object class in the historical remote sensing image;
dividing the probability of each ground object type in the remote sensing image and the historical remote sensing image which are taken as the sample area into a training sample and a testing sample according to a set proportion;
taking the historical remote sensing image in the training sample as the input of the deep learning model, and taking the probability of each ground object class appearing in the historical remote sensing image as the output of the deep learning model to train to obtain a trained deep learning model;
and verifying the trained deep learning model based on the historical remote sensing image in the test sample and the probability of each ground object class in the historical remote sensing image.
Preferably, the inputting the high-resolution remote sensing image into a pre-trained deep learning model to generate a class probability map of the surrounding environment of the power transmission channel includes:
performing convolution/pooling operation on the high-resolution remote sensing image to generate an image characteristic diagram;
inputting the image characteristic diagram and the high-resolution remote sensing image into a convolutional layer, and extracting and performing convolution calculation on the image characteristic diagram through the convolutional layer to obtain an image characteristic diagram array;
activating the image feature map array by using an activation function, inputting the image feature map array into a pooling layer, and performing pooling operation by using a maximum pooling method to obtain an image two-dimensional feature map;
inputting the image two-dimensional feature map into a full-connection layer, optimizing and mapping the image two-dimensional feature map, and outputting a one-dimensional feature map to a random forest classifier;
and classifying the one-dimensional characteristic graph by using the random forest classifier, and outputting a probability graph of the surrounding environment of the power transmission channel displayed in a grid form.
Preferably, the verifying the trained deep learning model based on the test sample includes:
inputting the historical remote sensing image of the surrounding environment of the power transmission channel in the test sample into the trained deep learning model to obtain a class probability map of the surrounding environment of the power transmission channel;
assigning corresponding category information to each object on the category probability map by adopting the difference degree based on the category probability map of the surrounding environment of the power transmission channel;
determining the probability of extracting the ground object based on the category information and the probability of each ground object category appearing in the historical remote sensing image in the test sample;
and when the probability reaches a set threshold value, determining the type of the ground feature, otherwise, performing sample selection and deep learning model training again until the maximum cycle value is reached.
Preferably, when the cycle maximum value is reached, if an object with a probability value smaller than a set threshold exists, the object is continuously segmented in a human-computer interaction mode;
and endowing the segmented objects with corresponding category information.
Preferably, the method further comprises deriving a probability of occurrence of each type of ground object in the historical remote sensing image as a vector result, and smoothing each object by using vector smoothing software.
Preferably, the multi-scale segmentation is performed on the remote sensing image of the surrounding environment of the power transmission channel by using the set scale to obtain a multi-region remote sensing image, and the multi-region remote sensing image comprises:
any single-pixel object in the high-resolution remote sensing image is used as a growth starting point;
merging the single-pixel object as a growth starting point and the single-pixel object with similar or same properties to the single-pixel object into a region where the growth starting point is located to form a region with a new object;
and repeating the steps by using the new object as a growth starting point until no object with the same or similar properties as the object as the growth starting point exists.
A deep learning-based power transmission channel surrounding environment classification system, comprising:
the information acquisition module is used for acquiring a high-resolution remote sensing image of the surrounding environment of the power transmission channel to be classified;
the class probability generation module is used for inputting the high-resolution remote sensing image into a pre-trained deep learning model to generate a class probability map of the surrounding environment of the power transmission channel;
the classification module is used for classifying the surrounding environment of the power transmission channel based on the class probability map to obtain the probability of each ground object class in the historical remote sensing image;
the deep learning model is obtained by training by using an object-oriented classification method based on the historical remote sensing images and the probability of each ground object type in the historical remote sensing images.
Preferably, the deep learning model training device further comprises a model training module, wherein the model training module is used for training the deep learning model by using an object-oriented classification method based on the historical remote sensing images and the probability of each terrain category appearing in the historical remote sensing images.
Preferably, the model training module includes:
the image acquisition submodule is used for acquiring a historical remote sensing image of the surrounding environment of the power transmission channel;
the scale division submodule is used for carrying out multi-scale division on the remote sensing image of the surrounding environment of the power transmission channel by utilizing a set scale to obtain a multi-region remote sensing image;
the sample area selection submodule is used for selecting a regional remote sensing image containing all ground feature categories from the multi-region remote sensing image as a sample area;
the classification submodule is used for carrying out full-element classification on the sample area by using an object-oriented classification method to obtain the probability of each ground object class in the historical remote sensing image;
the sample division submodule is used for dividing the probability of each ground object type in the remote sensing image and the historical remote sensing image which are taken as the sample area into a training sample and a testing sample according to a set proportion;
the training submodule is used for taking the historical remote sensing image in the training sample as the input of the deep learning model, and taking the probability of each ground object type appearing in the historical remote sensing image as the output of the deep learning model to train so as to obtain the trained deep learning model;
and the verification submodule verifies the trained deep learning model based on the historical remote sensing image in the test sample and the probability of each ground object type in the historical remote sensing image.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a method and a system for classifying surrounding environments of a power transmission channel based on deep learning, wherein the method and the system comprise the following steps: acquiring a high-resolution remote sensing image of the surrounding environment of a power transmission channel to be classified; inputting the high-resolution remote sensing image into a pre-trained deep learning model to generate a class probability map of the surrounding environment of the power transmission channel; classifying the power transmission channel surrounding environment based on the class probability map; the deep learning model is obtained by training by using an object-oriented classification method based on the historical remote sensing images and the probability of each ground object type in the historical remote sensing images. The classification of the surrounding environment of the power transmission channel is realized by utilizing the deep learning model and the object-oriented method, the high-resolution remote sensing image can be further utilized to classify the ground surface covering objects around the power transmission channel, and meanwhile, basic data support and auxiliary decision basis are provided for the safe operation of power facilities, emergency rescue planning and the like.
Drawings
FIG. 1 is a schematic diagram of a method for classifying the surrounding environment of a power transmission channel according to the present invention;
FIG. 2 is a flow chart of the power transmission channel environment classification operation of the present invention;
FIG. 3 is a schematic diagram illustrating the basic operation of the deep learning model of the present invention;
FIG. 4 is a schematic diagram of the deep learning model convolution operation of the present invention;
FIG. 5 is a diagram illustrating a deep learning model ReLU function representation according to the present invention;
FIG. 6 is a diagram of the deep learning model pooling operation of the present invention;
FIG. 7 is a schematic diagram of a fully connected layer structure of the deep learning model of the present invention.
Detailed Description
For a better understanding of the present invention, reference is made to the following description taken in conjunction with the accompanying drawings and examples.
The invention provides a deep learning-based power transmission channel surrounding environment classification method and system, and mainly relates to three main aspects of an image multi-scale segmentation technology, an object-oriented classification technology, a deep learning technology and the like, wherein the flow chart of the method is shown in figure 1, and the specific classification work flow chart is shown in figure 2.
Example 1
Firstly, obtaining a high-resolution remote sensing image to be classified
Second, make the training sample of the deep learning model
1. Firstly, carrying out multi-scale segmentation on the obtained historical remote sensing image.
In this embodiment, the high-resolution remote sensing image is segmented by selecting the scale 55 through repeated attempts on the input historical remote sensing image.
The adopted multi-scale segmentation is essentially the process of dividing a historical remote sensing image with M x N lines into a plurality of complementary overlapping regions (objects).
The multi-scale division adopts a region merging algorithm with minimum heterogeneity, the merging of pixels in the image division starts from any one pixel in an image, single pixels are merged into a smaller object firstly, then the smaller object is merged into a larger object, the heterogeneity of the object which continuously grows in the division process is minimum, and the process is from bottom to top and is a process of merging step by step.
The basic idea of the region merging method is to assemble regions with similar properties to form a region polygon, find a seed pixel as a starting point for growth for each region to be divided, merge pixels with the same or similar properties as the seed pixels in the neighborhood around the seed pixels into the region where the seed pixels are located, and continue the above process with the new pixels as new seed pixels until no pixel meeting the conditions exists, thereby generating an object.
The heterogeneity f of the object is determined by the spectral and geometric differences, the calculation formula:
f＝w·hcolor+(1-w)hshape
in the formula: h iscolorFor spectral heterogeneity, hshapeThe shape heterogeneity, w is weight, w is more than or equal to 0 and less than or equal to 1.
Spectral heterogeneity hcolorCalculated from the following equation:
in the formula: sigmaiIs the standard deviation, p, of the spectral value of the ith image layeriIs the weight of the ith image layer.
Shape heterogeneity hshapeCalculated from the following equation:
hshape＝w·hcmpct+(1-w)·hsmooth
in the formula: h issmoothSmoothness of image region boundary hcompctW is a weight, and w is more than or equal to 0 and less than or equal to 1.
hmoothAnd hcmpctDepending on the number of pixels N that make up the object, the actual side length L of the object is the side length M of the rectangle circumscribed by the object.
The basic unit of the history remote sensing image after multi-scale segmentation is an object formed by changing a single pixel into a homogeneous pixel, the multi-scale segmentation not only generates a meaningful image object, but also expands the object information of the history remote sensing image to different scales, and the multi-scale description of the remote sensing image information is realized.
2. And carrying out all-element classification on the sample area selected after the segmentation according to the set scale by using an object-oriented classification method, wherein the specific implementation steps are as follows:
and establishing a corresponding classification system according to the classification rules of different ground objects. Selecting partial areas on the historical remote sensing image as sample areas, and carrying out full-element classification on the sample areas by using an object-oriented classification method, wherein the principle of sample area selection is that all ground object types to be extracted need to be contained in the areas, and meanwhile, each ground object type has enough object quantity, and the probability of each ground object type appearing in the historical remote sensing image is obtained through the method and is used as output in subsequent deep learning model training.
3. And taking the obtained historical remote sensing image and the probability of each ground object type in the obtained historical remote sensing image as a sample for training the deep learning model, and dividing the sample for training the deep learning model into a training sample and a test sample according to a set proportion. In this embodiment, a certain amount of samples for training the deep learning model are set, and 80% of the samples are extracted as training samples, and the other 20% are extracted as test samples, so that in the subsequent model training, the training of the model can be performed by using the training samples, and then the training degree of the deep learning model is verified by using the test samples.
Thirdly, training the deep learning model
Deep learning is a new area in machine learning research. The main reason why deep learning has made great progress in the field of image recognition over conventional methods is that deep Convolutional Neural Networks (CNNs) automatically learn features through training data. In object-oriented classification, the deep learning technology is utilized to train the characteristics of the object such as spectrum and shape, and automatically train and master the high-level semantic characteristics of different objects such as texture and environment, and the method has incomparable advantages in remote sensing image classification.
In the implementation case, the deep learning model is trained by using the training samples, that is, the obtained historical remote sensing image is used as the input of the deep learning model training, the probability of each ground object type appearing in the historical remote sensing image is used as the output of the deep learning model, and the high-level semantic characteristics of the shape, texture, environment and the like of different objects in the probability of each ground object type appearing in the historical remote sensing image are obtained and used for guiding all subsequent object classifications of the deep learning model, so that the problem of accurately extracting the ground object information of the surrounding environment of the power transmission channel is effectively solved.
In the embodiment, the Caffe framework used by the deep learning model is a deep learning framework with excellent readability, simplicity and performance. The AlexNet model used is a typical convolutional neural network model, and mainly comprises the following basic operations: convolution, activation function, pooling, full join, and classifier. As shown in fig. 3.
1. Convolution with a bit line
The convolutional layer is used as the core of the neural network, and is very suitable for extracting high-level features with the help of strategies such as local perception, weight sharing and the like. The input sources of the convolution layer comprise historical remote sensing images and feature maps generated after convolution/pooling operation, and the size of the convolution kernel directly influences the refinement processing degree and the abstract effect of the images. In this embodiment, after repeated training, the convolution kernel of 11 × 11 size is selected to perform convolution operation on the input image, and the principle of the convolution operation is shown in fig. 4.
2. Activating a function
In the back propagation process of AlexNet training, because a chain rule of matrix derivation is used, if the number of successive multiplication of each layer is less than 1, the gradient is multiplied more forwards and less, and finally the gradient disappears. In this patent, a ReLU (Rectified Linear Unit) function is selected as an activation function to prevent gradient disappearance and also to further enhance the expression capability of the network. The representation of the ReLU function is shown in fig. 5.
3. Pooling
The feature dimensionality obtained by the convolution operation is high, and an overfitting phenomenon easily occurs in the process of training the classifier. The pooling operation is a process of dimension reduction and abstraction of the convolution layer output result, and can obtain low-dimensional feature representation. Through pooling, the purpose of describing a large-size feature map by using a small-size feature map can be achieved, the performance of the classifier is effectively improved, and overfitting is prevented. In this embodiment, the pooling layer calculation function selects the most common maximum pooling, and the principle is shown in fig. 6.
4. Full connection
After the input image is subjected to convolution-pooling for multiple times, the output result is a two-dimensional feature map, and the high-level features with high abstraction cannot be directly classified by using a traditional classifier. The structure of the fully-connected layer of the convolutional neural network is similar to that of the traditional neural network and is used for optimizing the characteristics of the convolutional-pooling processing output. In order to meet the classification requirement, the features extracted by the convolutional neural network are input into a full-connection layer network structure to realize the mapping and optimization from two-dimensional features to one-dimensional features. Finally, the one-dimensional features output by the fully connected layer are output to a classifier to obtain a final classification result, and the fully connected layer structure is shown in fig. 7.
The calculation formula of the full connection layer is as follows:
yi＝wi·yi-1+bi
in the formula: y isiIs the output vector of the ith fully-connected layer, yi-1Feature vector, w, output for i-1 fully-connected layersiAnd biThe weight parameter and offset of the ith fully-connected layer.
5. Classifier
The conventional Logistic classifier and Softmax classifier in deep learning can well solve the problem of image recognition. But for the areas with complicated ground features and easy confusion, the identification accuracy is not high. In the embodiment, a large number of experiments show that the Random Forest (RF) method in the traditional machine learning has a high prediction accuracy and is not easy to over-fit. Therefore, in the present embodiment, deep learning is combined with an RF classifier in conventional machine learning to classify the feature images obtained by training the AlexNet model, and a probability map of each feature class is generated.
The probability map is displayed by converting probability values of the object in the image belonging to the current surface feature classification into grid values of 0 to 255. For example, if the probability that a certain object belongs to a building, a forest land, a water body, and an artificial earth surface is 95%, 2%, and 1% through an AlexNet model and RF prediction, the grid value of the current object in the probability map of the building is 255 × 95% — 242, the grid value in the probability map of the forest land is 255 × 2% — 5, the grid value in the probability map of the water body is 255 × 2% — 5, and the grid value in the probability map of the road is 255 × 1% — 3.
Checking of deep learning model classification
Inputting the historical remote sensing image of the surrounding environment of the power transmission channel in the test sample into a trained deep learning model to obtain a class probability map of the surrounding environment of the power transmission channel;
in the implementation case, a certain historical remote sensing image in the test sample is used as the input of the deep learning model, the deep learning model trained by the training submodule is used for calculating the ground feature class probability in the certain historical remote sensing image, and after the deep learning model is calculated, a ground feature class probability map for the input certain historical remote sensing image is output. And repeating the steps until the deep learning model outputs the class probability graphs corresponding to all the historical remote sensing images in the test sample.
Based on the obtained category probability graph, corresponding category information is given to each object on the category probability graph by adopting the difference degree;
determining the probability of extracting the ground object in a category probability map obtained by the deep learning model based on the category information and the probability of each ground object category appearing in the historical remote sensing image in the test sample;
and when the probability reaches a set threshold value and the type of the ground object can be determined, the result obtained by the calculation of the deep learning model and the performance of the deep learning model are proved to meet the requirements.
Otherwise, the sample selection and the deep learning model training are carried out again until the probability in the class probability graph output by the deep learning model meets the set threshold value and the ground object type can be determined.
Fifth, classification of surrounding environment of power transmission channel
And the obtained remote sensing image of the surrounding environment of the power transmission channel is used as the input of a trained deep learning model, and a probability map of the surrounding environment category of the power transmission channel is generated through the calculation of the trained deep learning model.
And aiming at the generated power transmission channel surrounding environment category probability map, corresponding category information is given to each object in the obtained category probability map by utilizing the difference degree and is extracted, and the probability of each ground object category appearing in the power transmission channel surrounding environment remote sensing image is generated.
Five, multi-scale object-oriented semi-automatic editing
After the classification extracts the objects in the probability map, if there are still not extracted objects, it proves that the segmentation scale in the multi-scale segmentation method needs to be modified, so that the objects which are not extracted are continuously segmented by adopting a human-computer interaction mode.
Then, the object divided by the man-machine interaction mode is given to the class information of the object.
Sixthly, vector smoothing
And deriving the probability of each ground object type in the final remote sensing image of the surrounding environment of the power transmission channel as a vector result.
In the embodiment, in the process of deriving the vector result, Adobe illustrator software is selected to smooth each object, so that the boundary of the object is more natural and smooth.
Example 2
A power transmission channel surrounding environment classification system based on deep learning mainly comprises an information acquisition module, a category probability generation module, a classification module and a model training module, and a specific work flow schematic diagram is shown in figure 2.
1. Information acquisition module
An information acquisition module in a power transmission channel surrounding environment classification system based on deep learning is used for acquiring a high-resolution remote sensing image of the power transmission channel surrounding environment as input of the classification system.
2. Model training module
In the class probability generation module, a deep learning model trained in advance needs to be used for generating a power transmission channel surrounding environment class probability map for the power transmission channel surrounding environment, so the deep learning model needs to be trained by the model training module.
The model training module comprises: the system comprises an image acquisition submodule, a scale division submodule, a sample area selection submodule, a classification submodule, a sample division submodule, a training submodule and a verification submodule.
An image acquisition submodule: and acquiring a historical high-resolution remote sensing image of the surrounding environment of the power transmission channel by using the image acquisition submodule.
A scale division submodule: the scale division submodule adopts a multi-scale division method to divide the historical high-resolution remote sensing image of the surrounding environment of the power transmission channel.
The multi-scale segmentation method adopted in the scale segmentation submodule is essentially a process of dividing a remote sensing image of M x N lines into a plurality of complementary overlapping regions (objects).
The multi-scale segmentation method adopts a region merging algorithm with minimum heterogeneity, merging of pixels in image segmentation starts from any one pixel in an image, single pixels are merged into smaller objects firstly, then the smaller objects are merged into larger objects, and the heterogeneity of the objects which continuously grows in the segmentation process is minimum, so that the process of merging step by step from bottom to top is realized.
The basic idea of the region merging method is to assemble regions with similar properties to form a region polygon, find a seed pixel as a starting point for growth for each region to be divided, merge pixels with the same or similar properties as the seed pixels in the neighborhood around the seed pixels into the region where the seed pixels are located, and continue the above process with the new pixels as new seed pixels until no pixel meeting the conditions exists, thereby generating an object.
The heterogeneity f of the object is determined by the spectral and geometric differences, the calculation formula:
f＝w·hcolor+(1-w)hshape
in the formula: h iscolorFor spectral heterogeneity, hshapeThe shape heterogeneity, w is weight, w is more than or equal to 0 and less than or equal to 1.
Spectral heterogeneity hcolorCalculated from the following equation:
in the formula: sigmaiIs the standard deviation, p, of the spectral value of the ith image layeriIs the weight of the ith image layer.
Shape heterogeneity hshapeCalculated from the following equation:
hshape＝w·hcmpct+(1-w)·hsmooth
in the formula: h issmoothSmoothness of image region boundary hcompctW is a weight, and w is more than or equal to 0 and less than or equal to 1.
hmoothAnd hcmpctDepending on the number of pixels N that make up the object, the actual side length L of the object is the side length M of the rectangle circumscribed by the object.
The basic unit of the remote sensing image after the multi-scale segmentation is changed into an object consisting of homogeneous pixels from a single pixel, the multi-scale segmentation not only generates a meaningful image object, but also expands the information of the remote sensing image object to different scales, and the multi-scale description of the information of the remote sensing image is realized.
Therefore, in the scale division submodule, the input historical high-resolution remote sensing image is repeatedly tried for many times by adopting the multi-scale division method, and the scale 55 is selected to divide the high-resolution remote sensing image to obtain the multi-region remote sensing image.
A sample area selection submodule: and selecting a region with certain conditions from the multi-region remote sensing image obtained from the scale division submodule as a sample region.
In the process of selecting the sample area, the sample area selection submodule should select the sample area that satisfies the following conditions: firstly, the requirement in the region includes all the ground feature categories to be extracted; second, the region contains a sufficient number of objects in all the terrain categories to be extracted.
If a certain area in the multi-area remote sensing image meets the two conditions, the area is selected as a sample area by the sample area selection submodule.
A classification submodule: and aiming at the sample areas selected by the sample area selection submodule, carrying out full-factor classification on the sample areas by using an object-oriented classification method to obtain the probability of each ground object class in the historical remote sensing image.
Before carrying out full-element classification on the sample area, the classification submodule needs to establish a corresponding classification system according to different ground feature classification rules, and then carries out full-element classification on the sample area after the classification system is established.
A sample division submodule: and taking the historical remote sensing image of the surrounding environment of the power transmission channel acquired in the image acquisition submodule and the probability of occurrence of each ground object type in the historical remote sensing image acquired in the classification submodule as a model training sample, and dividing the model training sample into a training sample and a test sample according to a set proportion by the sample division submodule.
In this embodiment, the ratio is set to 4:1, and assuming that 100 model training samples are given, the sample division submodule divides 80 training samples from 100 model training samples and divides 20 training samples into 20 testing samples.
Training a submodule: the deep learning method comprises the following steps of taking the historical remote sensing image in the training sample divided from the sample submodule as the input of the deep learning model, taking the probability of each ground object type appearing in the historical remote sensing image as the output of the deep learning model, and training to obtain the trained deep learning model.
In the implementation case, the deep learning model is trained by the training submodule through the training sample, namely, the obtained remote sensing image is used as the input of the deep learning model training, the probability of each ground object type appearing in the historical remote sensing image is used as the output of the deep learning model, and the high-level semantic characteristics of the shape, texture, environment and the like of different objects in the probability of each ground object type appearing in the historical remote sensing image are obtained and used for guiding all subsequent object classifications of the deep learning model, so that the problem of accurately extracting the ground object information of the surrounding environment of the power transmission channel is effectively solved.
In the embodiment, the Caffe framework adopted by the training submodule is a deep learning framework with excellent readability, simplicity and performance. The AlexNet model used is a typical convolutional neural network model, and mainly comprises the following basic operations: convolution, activation function, pooling, full join, and classifier. As shown in fig. 3.
Convolution with a bit line
The convolutional layer is used as the core of the neural network, and is very suitable for extracting high-level features with the help of strategies such as local perception, weight sharing and the like. The input sources of the convolution layer comprise historical remote sensing images and feature maps generated after convolution/pooling operation, and the size of the convolution kernel directly influences the refinement processing degree and the abstract effect of the images. In this embodiment, after repeated training, the convolution kernel of 11 × 11 size is selected to perform convolution operation on the input image, and the principle of the convolution operation is shown in fig. 4.
Activating a function
In the back propagation process of AlexNet training, because a chain rule of matrix derivation is used, if the number of successive multiplication of each layer is less than 1, the gradient is multiplied more forwards and less, and finally the gradient disappears. In this patent, a ReLU (Rectified Linear Unit) function is selected as an activation function to prevent gradient disappearance and also to further enhance the expression capability of the network. The representation of the ReLU function is shown in fig. 5.
Pooling
The feature dimensionality obtained by the convolution operation is high, and an overfitting phenomenon easily occurs in the process of training the classifier. The pooling operation is a process of dimension reduction and abstraction of the convolution layer output result, and can obtain low-dimensional feature representation. Through pooling, the purpose of describing a large-size feature map by using a small-size feature map can be achieved, the performance of the classifier is effectively improved, and overfitting is prevented. In this embodiment, the pooling layer calculation function selects the most common maximum pooling, and the principle is shown in fig. 6.
Full connection
After the input image is subjected to convolution-pooling for multiple times, the output result is a two-dimensional feature map, and the high-level features with high abstraction cannot be directly classified by using a traditional classifier. The structure of the fully-connected layer of the convolutional neural network is similar to that of the traditional neural network and is used for optimizing the characteristics of the convolutional-pooling processing output. In order to meet the classification requirement, the features extracted by the convolutional neural network are input into a full-connection layer network structure to realize the mapping and optimization from two-dimensional features to one-dimensional features. Finally, the one-dimensional features output by the fully connected layer are output to a classifier to obtain a final classification result, and the fully connected layer structure is shown in fig. 7.
The calculation formula of the full connection layer is as follows:
yi＝wi·yi-1+bi
in the formula: y isiIs the output vector of the ith fully-connected layer, yi-1Feature vector, w, output for i-1 fully-connected layersiAnd biThe weight parameter and offset of the ith fully-connected layer.
Classifier
The conventional Logistic classifier and Softmax classifier in deep learning can well solve the problem of image recognition. But for the areas with complicated ground features and easy confusion, the identification accuracy is not high. In the embodiment, a large number of experiments show that the Random Forest (RF) method in the traditional machine learning has a high prediction accuracy and is not easy to over-fit. Therefore, in the present embodiment, deep learning is combined with an RF classifier in conventional machine learning to classify the feature images obtained by training the AlexNet model, and a probability map of each feature class is generated.
The probability map is displayed by converting probability values of the object in the image belonging to the current surface feature classification into grid values of 0 to 255. For example, if the probability that a certain object belongs to a building, a forest land, a water body, and an artificial earth surface is 95%, 2%, and 1% through an AlexNet model and RF prediction, the grid value of the current object in the probability map of the building is 255 × 95% — 242, the grid value in the probability map of the forest land is 255 × 2% — 5, the grid value in the probability map of the water body is 255 × 2% — 5, and the grid value in the probability map of the road is 255 × 1% — 3.
A checking submodule: the checking submodule checks the trained deep learning model according to the historical remote sensing image in the test sample and the probability of each ground object type in the historical remote sensing image, and the checking steps are as follows:
in the implementation case, a certain historical remote sensing image in the test sample is used as the input of the deep learning model, the deep learning model trained by the training submodule is used for calculating the ground feature class probability in the certain historical remote sensing image, and after the deep learning model is calculated, a ground feature class probability map for the input certain historical remote sensing image is output. And repeating the steps until the deep learning model outputs the class probability graphs corresponding to all the historical remote sensing images in the test sample.
Based on the obtained category probability graph, corresponding category information is given to each object on the category probability graph by adopting the difference degree;
determining the probability of extracting the ground object in a category probability map obtained by the deep learning model based on the category information and the probability of each ground object category appearing in the historical remote sensing image in the test sample;
and when the probability reaches a set threshold value and the type of the ground object can be determined, the result obtained by the calculation of the deep learning model and the performance of the deep learning model are proved to meet the requirements.
Otherwise, the sample selection and the deep learning model training are required to be carried out again until the probability in the class probability graph output by the deep learning model meets the set threshold value and the ground object type can be determined.
2. Category probability generation module
The category probability generation module comprises a deep learning module trained in the model training module, and the obtained remote sensing image of the surrounding environment of the power transmission channel is calculated by using the trained deep learning module to obtain a category probability map of the surrounding environment of the power transmission channel.
In this embodiment, the architecture adopted by the deep learning model in the category probability generation module and the calculation method of the deep learning model are the same as those in the model training module. And calculating to obtain a power transmission channel surrounding environment category probability map by utilizing a caffe framework and an Alexnet model through convolution, an activation function, pooling, full connection and a classifier.
3. Classification module
And aiming at the power transmission channel surrounding environment category probability map generated in the category probability generation module, a classification module is utilized to find out proper difference degree, corresponding category information is given to each object in the obtained category probability map, the category information is extracted, and the probability of each ground object category appearing in the remote sensing image around the power transmission channel is generated.
If the classification extracts the objects in the probability map, and if the objects are still not extracted, it proves that the segmentation scales in the scale segmentation sub-module need to be modified, so that the objects which are extracted are continuously segmented by adopting a human-computer interaction mode.
Then, the object divided by the man-machine interaction mode is given to the class information of the object.
And finally, deriving the probability of each ground object type in the final remote sensing image of the surrounding environment of the power transmission channel as a vector result.
In the implementation case, in the process of deriving the vector result, Adobe illustrator software is selected to smooth each vector result, so that the boundary of the obtained vector result is more natural and smooth.
Example 3
In the implementation case, the high-resolution remote sensing image of the surrounding environment of the power transmission channel in the area of the dense channel of Jia lake Zhejiang is used as the input of the deep learning model, and the pre-trained deep learning model is used for classifying the surrounding environment of the power transmission channel.
And after the input high-resolution remote sensing image is subjected to convolution, pooling, full connection and classifier operation in the deep learning model, outputting the high-resolution remote sensing image as a class probability map of the surrounding environment of the power transmission channel.
In order to ensure the accuracy of deep learning model classification, corresponding class information is continuously and accurately given to each object on a class probability map output by the deep learning model by using the difference degree;
determining the probability of extracting the ground object based on the class information and the probability of each ground object class appearing in the power transmission channel surrounding environment class probability graph;
and when the probability reaches a set threshold value, determining the probability of the ground feature type, otherwise, performing sample selection and deep learning model training again until the maximum cycle value is reached.
When the cycle maximum value is reached, if an object with the probability value smaller than a set threshold value exists, continuously segmenting the object by adopting a human-computer interaction mode; and endowing the segmented objects with corresponding category information.
The probability of each ground object type appearing in the finally obtained remote sensing image of the surrounding environment of the power transmission channel is exported to be a vector result, and each object is subjected to smoothing processing in the process of exporting to be the vector result, so that the boundary of the object is smoother
By using the method to classify the high-resolution satellite remote sensing images of the surrounding environment of the power transmission channel in the Jia lake region in Zhejiang, the environment land feature information such as artificial earth surface, buildings, water, forest land and farmland (including vegetation farmland and non-vegetation farmland) around the power transmission channel can be classified, and a good classification effect is obtained
It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The present invention is not limited to the above embodiments, and any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are included in the scope of the claims of the present invention which are filed as the application.
Claims (10)
1. A method for classifying surrounding environment of a power transmission channel based on deep learning is characterized by comprising the following steps:
acquiring a high-resolution remote sensing image of the surrounding environment of a power transmission channel to be classified;
inputting the high-resolution remote sensing image into a pre-trained deep learning model to generate a class probability map of the surrounding environment of the power transmission channel;
classifying the power transmission channel surrounding environment based on the class probability map;
the deep learning model is obtained by training by using an object-oriented classification method based on the historical remote sensing images and the probability of each ground object type in the historical remote sensing images.
2. The method of claim 1, wherein the training of the deep learning model comprises:
acquiring a historical remote sensing image of the surrounding environment of a power transmission channel;
carrying out multi-scale segmentation on the historical remote sensing image of the surrounding environment of the power transmission channel by using a set scale to obtain a multi-region remote sensing image;
selecting a regional remote sensing image containing all ground feature categories from the multi-region remote sensing image as a sample region;
carrying out full-element classification on the sample area by using an object-oriented classification method to determine the probability of each ground object class in the historical remote sensing image;
dividing the probability of each ground object type in the remote sensing image as the sample area and the historical remote sensing image of the historical remote sensing image into a training sample and a testing sample according to a set proportion;
taking the historical remote sensing image in the training sample as the input of the deep learning model, and taking the probability of each ground object class appearing in the historical remote sensing image as the output of the deep learning model to train to obtain a trained deep learning model;
and verifying the trained deep learning model based on the historical remote sensing image in the test sample and the probability of each ground object class in the historical remote sensing image.
3. The method of claim 2, wherein the step of inputting the high-resolution remote sensing image into a pre-trained deep learning model to generate a class probability map of the surrounding environment of the power transmission channel comprises the following steps:
performing convolution/pooling operation on the high-resolution remote sensing image to generate an image characteristic diagram;
inputting the image characteristic diagram and the high-resolution remote sensing image into a convolutional layer, and extracting and performing convolution calculation on the image characteristic diagram through the convolutional layer to obtain an image characteristic diagram array;
activating the image feature map array by using an activation function, inputting the image feature map array into a pooling layer, and performing pooling operation by using a maximum pooling method to obtain an image two-dimensional feature map;
inputting the image two-dimensional feature map into a full-connection layer, optimizing and mapping the image two-dimensional feature map, and outputting a one-dimensional feature map to a random forest classifier;
and classifying the one-dimensional characteristic graph by using the random forest classifier, and outputting a probability graph of the surrounding environment of the power transmission channel displayed in a grid form.
4. The method of claim 2, wherein verifying the trained deep learning model based on the test samples comprises:
inputting the historical remote sensing image of the surrounding environment of the power transmission channel in the test sample into the trained deep learning model to obtain a class probability map of the surrounding environment of the power transmission channel;
assigning corresponding category information to each object on the category probability map by adopting the difference degree based on the category probability map of the surrounding environment of the power transmission channel;
determining the probability of extracting the ground object based on the category information and the probability of each ground object category appearing in the historical remote sensing image in the test sample;
and when the probability reaches a set threshold value, determining the probability of the ground feature type, otherwise, performing sample selection and deep learning model training again until the maximum cycle value is reached.
5. The method of claim 4, further comprising, when the cycle maximum is reached, if there is an object with a probability value smaller than a set threshold, continuing to segment the object by human-computer interaction;
and endowing the segmented objects with corresponding category information.
6. The method of claim 4, further comprising deriving as vector results probabilities of occurrence of each of the terrain categories in the remote-sensed image of the environment surrounding the power transmission channel, and smoothing each of the vector results with vector smoothing software.
7. The method according to claim 2, wherein the multi-scale segmentation of the remote sensing image of the surrounding environment of the power transmission channel by using the set scale to obtain a multi-region remote sensing image comprises:
any single-pixel object in the historical remote sensing image is used as a growth starting point;
merging the single-pixel object as a growth starting point and the single-pixel object with similar or same properties to the single-pixel object into a region where the growth starting point is located to form a region with a new object;
and repeating the steps by using the new object as a growth starting point until no object with the same or similar properties as the object as the growth starting point exists.
8. A power transmission channel surrounding environment classification system based on deep learning is characterized by comprising:
the information acquisition module is used for acquiring a high-resolution remote sensing image of the surrounding environment of the power transmission channel to be classified;
the class probability generation module is used for inputting the high-resolution remote sensing image into a pre-trained deep learning model to generate a class probability map of the surrounding environment of the power transmission channel;
the classification module is used for classifying the surrounding environment of the power transmission channel based on the class probability map to obtain the probability of each ground object class in the surrounding environment of the power transmission channel;
the deep learning model is obtained by training historical remote sensing images as training samples and the probability of each ground object class in the historical remote sensing images obtained by carrying out full-element classification on the remote sensing images by using an object-oriented classification method.
9. The system of claim 8, further comprising a model training module for training the deep learning model by an object-oriented classification method based on the historical remote sensing images and the probability of each terrain category appearing in the historical remote sensing images.
10. The system of claim 8, wherein the model training module comprises:
the image acquisition submodule is used for acquiring a historical remote sensing image of the surrounding environment of the power transmission channel;
the scale division submodule is used for carrying out multi-scale division on the remote sensing image of the surrounding environment of the power transmission channel by utilizing a set scale to obtain a multi-region remote sensing image;
the sample area selection submodule is used for selecting a regional remote sensing image containing all ground feature categories from the multi-region remote sensing image as a sample area;
the classification submodule is used for carrying out full-element classification on the sample area by using an object-oriented classification method to obtain the probability of each ground object class in the historical remote sensing image;
the sample division submodule is used for dividing the probability of each ground object type in the remote sensing image and the historical remote sensing image which are taken as the sample area into a training sample and a testing sample according to a set proportion;
the training submodule is used for taking the historical remote sensing image in the training sample as the input of the deep learning model, and taking the probability of each ground object type appearing in the historical remote sensing image as the output of the deep learning model to train so as to obtain the trained deep learning model;
and the verification submodule verifies the trained deep learning model based on the historical remote sensing image in the test sample and the probability of each ground object type in the historical remote sensing image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010960023.2A CN112257496A (en) | 2020-09-14 | 2020-09-14 | Deep learning-based power transmission channel surrounding environment classification method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010960023.2A CN112257496A (en) | 2020-09-14 | 2020-09-14 | Deep learning-based power transmission channel surrounding environment classification method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112257496A true CN112257496A (en) | 2021-01-22 |
Family
ID=74232835
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010960023.2A Pending CN112257496A (en) | 2020-09-14 | 2020-09-14 | Deep learning-based power transmission channel surrounding environment classification method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112257496A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113160183A (en) * | 2021-04-26 | 2021-07-23 | 山东深蓝智谱数字科技有限公司 | Hyperspectral data processing method, device and medium |
| CN113723464A (en) * | 2021-08-02 | 2021-11-30 | 北京大学 | Remote sensing image classification method and device |
| CN115457388A (en) * | 2022-09-06 | 2022-12-09 | 湖南经研电力设计有限公司 | Power transmission and transformation remote sensing image ground feature identification method and system based on deep learning optimization |
-
2020
- 2020-09-14 CN CN202010960023.2A patent/CN112257496A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113160183A (en) * | 2021-04-26 | 2021-07-23 | 山东深蓝智谱数字科技有限公司 | Hyperspectral data processing method, device and medium |
| CN113723464A (en) * | 2021-08-02 | 2021-11-30 | 北京大学 | Remote sensing image classification method and device |
| CN113723464B (en) * | 2021-08-02 | 2023-10-03 | 北京大学 | Remote sensing image classification method and device |
| CN115457388A (en) * | 2022-09-06 | 2022-12-09 | 湖南经研电力设计有限公司 | Power transmission and transformation remote sensing image ground feature identification method and system based on deep learning optimization |
| CN115457388B (en) * | 2022-09-06 | 2023-07-28 | 湖南经研电力设计有限公司 | Power transmission and transformation remote sensing image ground object identification method and system based on deep learning optimization |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111598174B (en) | Model training method based on semi-supervised antagonistic learning and image change analysis method | |
| CN103049763B (en) | Context-constraint-based target identification method | |
| WO2023029461A1 (en) | Massive high-dimensional ais trajectory data clustering method | |
| CN109447994A (en) | In conjunction with the remote sensing image segmentation method of complete residual error and Fusion Features | |
| CN112257496A (en) | Deep learning-based power transmission channel surrounding environment classification method and system | |
| US11854244B2 (en) | Labeling techniques for a modified panoptic labeling neural network | |
| CN111259936B (en) | Image semantic segmentation method and system based on single pixel annotation | |
| CN110889449A (en) | Edge-enhanced multi-scale remote sensing image building semantic feature extraction method | |
| CN109409240A (en) | A kind of SegNet remote sensing images semantic segmentation method of combination random walk | |
| CN114694038A (en) | High-resolution remote sensing image classification method and system based on deep learning | |
| CN110084817A (en) | Digital elevation model production method based on deep learning | |
| CN113256649B (en) | Remote sensing image station selection and line selection semantic segmentation method based on deep learning | |
| CN112419333B (en) | Remote sensing image self-adaptive feature selection segmentation method and system | |
| CN112950780B (en) | Intelligent network map generation method and system based on remote sensing image | |
| CN113988147B (en) | Multi-label classification method and device for remote sensing image scene based on graph network, and multi-label retrieval method and device | |
| CN113591617B (en) | Deep learning-based water surface small target detection and classification method | |
| CN104408733A (en) | Object random walk-based visual saliency detection method and system for remote sensing image | |
| Feng et al. | Embranchment cnn based local climate zone classification using sar and multispectral remote sensing data | |
| CN112001293A (en) | Remote sensing image ground object classification method combining multi-scale information and coding and decoding network | |
| CN114943876A (en) | Cloud and cloud shadow detection method and device for multi-level semantic fusion and storage medium | |
| CN110807485A (en) | Method for fusing two-classification semantic segmentation maps into multi-classification semantic map based on high-resolution remote sensing image | |
| Louw et al. | Object-based land surface segmentation scale optimisation: An ill-structured problem | |
| Xiao et al. | Open-pit mine road extraction from high-resolution remote sensing images using RATT-UNet | |
| CN104036294A (en) | Spectral tag based adaptive multi-spectral remote sensing image classification method | |
| EP4174792A1 (en) | Method for scene understanding and semantic analysis of objects |
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 |