CN110879992A - Grassland surface covering object classification method and system based on transfer learning - Google Patents
Grassland surface covering object classification method and system based on transfer learning Download PDFInfo
- Publication number
- CN110879992A CN110879992A CN201911183095.4A CN201911183095A CN110879992A CN 110879992 A CN110879992 A CN 110879992A CN 201911183095 A CN201911183095 A CN 201911183095A CN 110879992 A CN110879992 A CN 110879992A
- Authority
- CN
- China
- Prior art keywords
- grassland
- remote sensing
- sensing image
- image
- surface covering
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Abstract
The application relates to a grassland surface covering classification method, in particular to a grassland surface covering classification method and system based on transfer learning. The method of the present application comprises: obtaining a grassland remote sensing image of a target area; preprocessing the remote sensing image of the grassland to obtain preprocessed remote sensing image data of the grassland; inputting the preprocessed grassland remote sensing image data into a trained grassland remote sensing image classification model, and outputting an image category; wherein the image category is one of preset earth surface covering categories. According to the method, the processed remote sensing image training model of the grassland can be directly input without performing work such as feature point marking, optimal feature combination and extraction on the grassland image, and the trained remote sensing image classification model of the grassland is used for classifying and identifying the remote sensing image of the grassland; the classification result is analyzed, the range change condition of various earth surface coverings can be obtained every year, and technical support is provided for the understanding of ecological conditions of related departments.
Description
Technical Field
The application relates to the field of grassland remote sensing image processing, in particular to a grassland surface covering classification method and system based on transfer learning.
Background
The grassland is the largest ecosystem of land, has important effects on the aspects of agriculture, animal husbandry, human survival and development and the like in the process of energy flow and material circulation, and the natural grassland resources in China are rich, the area of the natural grassland in China is about 3.928 hundred million hectares, and the area of the natural grassland in China occupies about 12 percent of the area of the grassland in the whole world, so that the protection and construction of the grassland are highly emphasized in China, the development of the grassland career is comprehensively promoted, and the investment of the grassland ecological construction project is increased. However, for a long time, the grassland ecosystem is influenced by both external factors (human factors) and internal factors (natural factors), so that the ecological degradation of the grassland and the reduction of the bearing capacity of the grassland are caused, and the dynamic monitoring and the comprehensive and accurate general investigation of the ecological conditions of the grassland are difficult.
The high-resolution remote sensing technology has the advantages of long time effect, high resolution, large breadth, high efficiency and the like. In recent years, with gradual emission operation of high-resolution satellites in China, high-resolution remote sensing images are gradually used as main data sources of remote sensing services, and a plurality of researchers combine the remote sensing images with deep learning and machine learning theories to perform image classification and target identification. Meanwhile, the remote sensing image provides a large amount of rapid and accurate data for various fields such as grassland classification, grassland dynamic monitoring, planning of national significant grassland ecological construction engineering, grassland resource and ecological research and the like.
At present, the classification method of grassland surface coverings mainly comprises the following steps: (1) a maximum likelihood classification method. Firstly, data preprocessing is carried out to enable surface feature information on an image to be highlighted, a combined waveband is determined, a true color image is simulated to be synthesized into an RGB false color image for display, the chromaticity can be fully corresponding to surface feature classification, representative data are selected from the image to be a training sample, a specific classifier is used for dividing a feature space based on a feature vector set, the probability that a pixel belongs to a certain class is calculated, the class is determined according to the number with the maximum probability, and classification is finally completed. Whether the spectral characteristic value of the image data obeys normal distribution or not during the maximum likelihood method classification has a large influence on the classification effect, the spectral characteristic of the greenbelt is closest to the normal distribution, the method is suitable for the maximum likelihood method classification method, and the classification effect of other earth surface covering objects is not ideal. (2) And (4) a decision tree classification algorithm. The method comprises the steps of carrying out conversion analysis on various characteristic variables on an original image, constructing an optimal classification characteristic variable combination wave band, selecting an optimal characteristic combination according to a statistical analysis result of field sample points on each characteristic variable, acquiring classification knowledge of related ground object types from a large amount of remote sensing data through data mining, finally formulating a ground cover classification rule and a threshold value thereof, operating the rule, and classifying. The decision tree classification algorithm needs a large amount of tests to obtain the optimal classification characteristic variable combination, the optimal characteristic combination is determined by combining ground monitoring, the determination is difficult, and the classification precision with similar extraction rules is low. (3) And (4) a deep confidence network classification algorithm. The method comprises the steps of combining spectrum-space two kinds of characteristic information, classifying the hyperspectral remote sensing images by using a depth confidence network, enabling spectrum vectors to be formed by spectra of all wave bands correspondingly, enabling the space vectors to use principal components obtained after principal component analysis, reducing data dimensionality of the spectrum characteristics, extracting neighborhood from a certain number of principal component components after principal component transformation is finished, constructing space information vectors, and classifying images by using deep features of data extracted by the depth confidence network. Although the deep confidence network model can well extract the characteristics of the image, the characteristics of the high-dimensional data image cannot be directly extracted, a large number of tests are also needed for selecting proper spatial information dimensions, the network structure and parameter selection of the model need to be optimized according to experience or experimental methods, and the operation time is long. (4) And (4) a convolutional neural network classification algorithm. The method comprises the steps of fusing single-source characteristics or spectral characteristics, textural characteristics, spatial structure characteristics and the like of remote sensing images in a vector or matrix form according to spatial dimensions, firstly transforming original data by utilizing principal component analysis, then extracting characteristic information by utilizing a convolution layer, taking a characteristic graph generated by convolution operation as input of a full connection layer, and finally finishing classification of the hyperspectral remote sensing images. In the classification of the high-spectrum remote sensing image, the high-spectrum image cannot be directly classified due to the integration of maps, high dimension and the existence of nonlinear components in data, and the image generally needs to be subjected to dimension reduction treatment to be classified.
Aiming at the problems of the grassland surface covering classification method, the application provides a grassland surface covering classification method and system based on transfer learning.
Disclosure of Invention
In order to solve the problems in the prior art, the migration learning algorithm is mainly applied to carry out classification research on grassland surface coverings through high-resolution remote sensing images, classification results are analyzed, the range change conditions of various surface coverings can be obtained, accurate, complete and rich scientific bases are provided for relevant departments, the utilization rate of grassland natural resources can be improved through implementation of relevant policies, effective implementation of relevant protection policies such as grassland ecological reward is guaranteed, reasonable utilization of grasslands is strengthened, ecological construction is strengthened, ecological rewarding is implemented, expected targets are achieved, and meanwhile execution cost and manpower liberation can be greatly reduced.
The application provides a grassland surface covering classification method based on transfer learning, which comprises the following steps: obtaining a grassland remote sensing image of a target area; preprocessing the grassland remote sensing image to obtain preprocessed grassland remote sensing image data; inputting the preprocessed grassland remote sensing image data into a trained grassland remote sensing image classification model, and outputting an image category; wherein the image category is one of preset earth surface covering categories.
In a preferred embodiment of the method for classifying grassland surface coverings, the method for preprocessing the grassland remote sensing image to obtain preprocessed grassland remote sensing image data includes: carrying out true color band synthesis on remote sensing image data of the grassland through corresponding software according to the sequence of red, green and blue (RGB); deriving the images synthesized by the wave bands into TIF format images according to a preset proportion; and cutting the derived TIF format image into a plurality of images with preset sizes to obtain preprocessed remote sensing image data of the grassland.
In a preferred embodiment of the above classification method for grassland surface coverings, the preset ratio is 1: 100000; and/or the predetermined size is 128 x 128; and/or the preset earth surface covering categories comprise: roads, grasses, sand, forests, and edges between grasses, sand, and forests.
In the preferred embodiment of the grassland surface covering object classification method, the training method of the grassland remote sensing image classification model comprises the steps of constructing a training data set; building a VGG network model based on transfer learning as a grassland remote sensing image classification model to be trained; and inputting the training data set into the grassland remote sensing image classification model to be trained so as to train the grassland remote sensing image classification model.
In a preferred embodiment of the above grassland surface covering classification method, the constructing a training data set includes: acquiring remote sensing image data of a grassland; carrying out true color band synthesis on remote sensing image data of the grassland through corresponding software according to the sequence of red, green and blue (RGB); deriving the images synthesized by the wave bands into TIF format images according to a preset proportion; cutting the derived TIF format image into a plurality of images with preset sizes; screening the cut image, deleting the white image, and reserving the image containing the image information; rotating the screened image for multiple times at different angles, and storing the rotated image for each time; and marking categories of the screened image data according to the preset earth surface covering categories, and dividing the image data after the categories are marked into a training set, a verification set and a test set to obtain a training data set meeting the requirements.
In an embodiment of the method for classifying grassland surface coverings, constructing a VGG network model based on transfer learning as a classification model of the remote sensing images of the grassland to be trained includes: reading a VGG network model which adopts a VGG19 model; and connecting two full-connection layers behind the VGG19 to obtain the grassland remote sensing image classification model to be trained.
In a preferred embodiment of the above grassland surface covering classification method, inputting a training data set into a grassland remote sensing image classification model to be trained to train the grassland remote sensing image classification model, the method includes: training a grassland remote sensing image classification model to be trained by taking training set data as a training sample; determining network structure and control parameters using the validation set data; after the grassland remote sensing image classification model is trained, the test set data is input into the grassland remote sensing image classification model to test whether the output image category meets the requirements.
The application also provides a grassland surface covering classification system based on transfer learning, and the grassland surface covering classification system comprises: the data acquisition module is used for acquiring a grassland remote sensing image of a target area; the preprocessing module is used for preprocessing the grassland remote sensing image to obtain preprocessed grassland remote sensing image data; the classification processing module comprises a trained grassland remote sensing image classification model and is configured to output an image category according to the preprocessed grassland remote sensing image data; wherein the image category is one of preset earth surface covering categories.
In an embodiment of the grassland remote sensing image classification system, the preprocessing module is specifically configured to: carrying out true color band synthesis on remote sensing image data of the grassland through corresponding software according to the sequence of red, green and blue (RGB); deriving the images synthesized by the wave bands into TIF format images according to a preset proportion; and cutting the derived TIF format image into a plurality of images with preset sizes to obtain preprocessed remote sensing image data of the grassland.
In a preferred embodiment of the grassland remote sensing image classification system, the preset ratio is 1: 100000; and/or the predetermined size is 128 x 128; and/or the preset earth surface covering categories comprise: roads, grasses, sand, forests, and edges between grasses, sand, and forests.
According to the grassland surface covering object classification method, a VGG network model based on transfer learning is built for training, the works of marking characteristic points, carrying out optimal characteristic combination, extracting and the like on grassland images are not needed, the processed grassland remote sensing image training model can be directly input, and the trained grassland remote sensing image classification model is used for carrying out classification and identification on the grassland remote sensing images; the classification results are analyzed, the range change conditions of various earth surface coverings every year can be obtained, technical support is provided for understanding of ecological conditions by relevant departments, accurate, complete and rich scientific bases are provided, the implementation of relevant policies can improve the utilization rate of grassland natural resources, the effective implementation of relevant protection policies such as grassland ecological reward and the like is guaranteed, reasonable utilization of grasslands is strengthened, ecological construction is enhanced, ecological reward is implemented, expected targets are achieved, and meanwhile the implementation cost and the liberation manpower can be greatly reduced.
Drawings
Fig. 1 is a main flowchart of a grassland surface covering classification method based on transfer learning according to an embodiment of the present application.
Fig. 2 is a main flowchart of a method for training a classification model of grassland surface coverings according to an embodiment of the present application.
Detailed Description
Various aspects and features of the present application are described herein with reference to the drawings.
It will be understood that various modifications may be made to the embodiments of the present application. Accordingly, the foregoing description should not be construed as limiting, but merely as exemplifications of embodiments. Those skilled in the art will envision other modifications within the scope and spirit of the application.
The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the application and, together with a general description of the application given above and the detailed description of the embodiments given below, serve to explain the principles of the application.
These and other characteristics of the present application will become apparent from the following description of preferred forms of embodiment, given as non-limiting examples, with reference to the attached drawings.
It should also be understood that, although the present application has been described with reference to some specific examples, a person of skill in the art shall certainly be able to achieve many other equivalent forms of application, having the characteristics as set forth in the claims and hence all coming within the field of protection defined thereby.
The above and other aspects, features and advantages of the present application will become more apparent in view of the following detailed description when taken in conjunction with the accompanying drawings.
Specific embodiments of the present application are described hereinafter with reference to the accompanying drawings; however, it is to be understood that the disclosed embodiments are merely exemplary of the application, which can be embodied in various forms. Well-known and/or repeated functions and constructions are not described in detail to avoid obscuring the application of unnecessary or unnecessary detail. Therefore, specific structural and functional details disclosed herein are not to be interpreted as limiting, but merely as a basis for the claims and as a representative basis for teaching one skilled in the art to variously employ the present application in virtually any appropriately detailed structure.
The specification may use the phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," which may each refer to one or more of the same or different embodiments in accordance with the application.
Referring to fig. 1, fig. 1 is a main flowchart of a grassland surface covering classification method based on transfer learning according to an embodiment of the present application. As shown in fig. 1, the grassland surface covering classification method of the present application includes: s110, obtaining a grassland remote sensing image of a target area; s120, preprocessing the remote sensing image of the grassland to obtain preprocessed remote sensing image data of the grassland; s130, inputting the preprocessed grassland remote sensing image data into a trained grassland remote sensing image classification model, and outputting an image category; wherein the image category is one of preset ground surface covering categories.
Specifically, in step S110, the target area is a grassland area of the grassland surface covering category to be identified, and the grassland remote sensing image is a high-resolution remote sensing image for the target area acquired by a high-resolution remote sensing technology. After the remote sensing image of the grassland of the target area is acquired, step S120 is executed.
In step S120, preprocessing the remote sensing image of the grassland, specifically including the following steps: and S210, carrying out true color band synthesis on the remote sensing image data of the grassland through corresponding software according to the sequence of red, green and blue (RGB). Wherein the software may be ArcGISI 10.2 software. And S220, deriving the images synthesized by the wave bands into TIF format images according to a preset proportion. Wherein, the preset ratio may be 1: 100000. other suitable ratios can be selected by those skilled in the art according to actual needs. And S230, cutting the derived TIF format image into a plurality of images with preset sizes to obtain preprocessed remote sensing image data of the grassland. As an example, the derived TIF format image may be cropped using MATLAB2016a, and the preset size of the cropping may be 128 × 128, or may be cropped to other sizes as needed. As will be understood by those skilled in the art, the grassland remote sensing image obtained in step S110 has a large coverage area, and multiple types of surface coverings are usually included in one grassland remote sensing image. The step is to cut the remote-sensing grassland image into a plurality of small-size images and identify the surface covering type in each small-size image, so that the surface covering type contained in the whole remote-sensing grassland image can be obtained. After the preprocessed remote sensing image data of the grassland is obtained, step S130 is executed.
In step S130, each of the preprocessed small-size images is input into a trained grassland remote sensing image classification model, and the trained grassland remote sensing image classification model outputs an image category, which is one of preset surface covering categories. That is, each image category represents a type of surface covering. Wherein, the person skilled in the art can classify the surface covering into different categories according to the actual need or purpose. In an embodiment of the present application, the preset ground surface covering categories include: roads (including residential roads and illegal roads), grasslands, sandy roads, forests, and edges between grasslands, sandy roads, and forests. Wherein the grassland is green or dark green; the sand is white or light yellow, irregular in shape and has obvious sand ripple; the forest land is dark and speckled, and has irregular geometric shape and uneven texture; the road (including residential areas, illegal roads and the like) has larger values of brightness and length-width ratio, reflects the degrees and the variation trends of grassland degeneration, land desertification, sand prevention and control and soil pollution in the current research area through the variation analysis of various quantities per year, and laterally reflects the implementation effects of ecological related policies such as grassland compensation, seal protection, sand prevention and control engineering, seal withdrawal and the like in the research area.
In the grassland surface covering classification method based on transfer learning provided by the application, step S130 is to identify the grassland remote sensing image by using the trained grassland remote sensing image classification model, wherein the training method of the grassland remote sensing image classification model comprises the following steps: s310, constructing a training data set; s320, building a VGG network model based on transfer learning as a remote sensing image classification model of the grassland to be trained; s330, inputting the training data set into the grassland remote sensing image classification model to be trained to train the grassland remote sensing image classification model.
Specifically, in step S310, a remote sensing data set of a certain grassland area is used as basic data, and then the grassland remote sensing image data is subjected to true color band synthesis through corresponding software (such as arcbios 10.2) according to the sequence of red, green and blue (RGB); then, deriving the image synthesized by the wave bands into a TIF format image according to a preset proportion (such as 1: 100000); the derived TIF format image is then cropped to a number of preset size images, which may be cropped to 128 x 128 size using MATLAB2016 a. Then screening the cut image, deleting the white image, and reserving the image containing the image information; and rotating the screened image for multiple times at different angles, and storing the rotated image for each time. For example, the training data set can be further expanded by rotating the screened image by 90 °, 180 °, and 270 ° with python3.6, and storing the rotated image. Marking the category (adopting the manual marking category) of the screened image data according to the preset ground surface covering category, in the embodiment of the application, the category of the mark is five categories, including: roads (including residential roads and illegal roads), grasslands, sandy roads, forests, and edges between grasslands, sandy roads, and forests. Further, the image data after the class marking is divided into a training set, a verification set and a test set, so that a training data set meeting requirements is obtained. As an example, the training set, the validation set, and the test set each contain these five categories of image data.
In step S320, a VGG network model based on transfer learning is built as a classification model of the remote sensing image of the grassland to be trained. As an example, a VGG network model based on migration learning is built under a Keras deep learning framework using python3.6 software. The VGG deep convolution network is used in the application, local receptive field is small, the filter of 3X3 size is used to whole convolution network, the network structure is deep, 19 layers exist, the model structure comprises 5 big convolution layers and the biggest pooling layer, and 3 big full-link layers F. Modifying an input 128 x 128 image into a size 224 x 224 specified by a network, using a prediction model of VGG19, taking a feature extraction part as a basis of transfer learning, reserving a network structure of 16 convolutional layers and a pooling layer, reserving training weights, modifying full-connection layer information, setting the number of full-connection layers to be 1400, selecting a Softmax activation function, adding a Dropout layer to prevent overfitting, and setting the classification number of the last full-connection layer to be 5 types. And setting parameters of an evaluation function, carrying out performance evaluation on the training model, and selecting an optimal model. Adam of the adaptive learning rate is selected by the optimization function, the initial learning rate is set to be 0.0001, and the accuracy of the network is improved. The loss function (loss) selects a cross entropy loss function (coordinated _ cross). The proximity of the actual output to the desired output is determined. The batch size is set to 64, i.e. 64 images are input for training at each training.
In step S330, the training data set is input into the grassland remote sensing image classification model to be trained to train the grassland remote sensing image classification model. The method specifically comprises the following steps: training a grassland remote sensing image classification model to be trained by taking training set data as a training sample; determining network structure and control parameters using the validation set data; after the grassland remote sensing image classification model is trained, the test set data is input into the grassland remote sensing image classification model to test whether the output image category meets the requirements. And after the grassland remote sensing image classification model is trained, storing the trained grassland remote sensing image classification model. And then, inputting the preprocessed grassland remote sensing image (cut into 128 × 128 images) into the trained grassland remote sensing image classification model, modifying the input 128 × 128 images into the size of 224 × 224 specified by the network, reading the trained grassland remote sensing image classification model, and outputting a classification result. And if the classification result is correct, applying the trained grassland remote sensing image classification model.
As described above, the method and the device for identifying the grassland surface covering by using the VGG19 network of transfer learning do not need to consider the characteristic of the features and the combination mode of the optimal classification features, and directly extract various features, so that the process of preferentially extracting the image features is omitted; the overfitting problem caused by the limited sample capacity can be effectively solved.
According to the grassland surface covering object classification method, the grassland images do not need to be subjected to work such as feature point labeling, optimal feature combination and extraction, the processed grassland remote sensing image training model can be directly input, and the trained grassland remote sensing image classification model is used for classifying and identifying the grassland remote sensing images; the classification result is analyzed, the range change condition of various earth surface coverings can be obtained every year, and technical support is provided for the understanding of ecological conditions of related departments. Specifically, the method comprises the following steps: (1) the grassland is mainly green or dark green, and the implementation effects of the grassland degradation degree, the water and soil loss and relevant policies (grassland compensation and the like) are known according to the annual variation of the total number of the grass land categories and the edge categories. (2) The sand land is mainly white or light yellow, irregular in shape and obvious in sand ripple, the desertification expansion condition is preliminarily analyzed and judged through the sand land classification quantity, edge data are analyzed, the condition that the flowing sand land is changed into a semi-fixed sand land and a fixed sand land is evaluated, and the effect of sand fixation is achieved. (3) The forest lands are dark and speckled, have irregular geometric shapes and uneven textures, and the quantity change of the forest lands is analyzed to reflect the effect of artificial afforestation and aerial seeding afforestation in sand prevention and control. (4) The constructed road and the illegal road have higher brightness and aspect ratio, and the change of the road to land utilization and soil environment is analyzed according to the change of the quantity of the road, so that the degradation of the marginal ecology of the road is caused, and the problems of desertification of the land around the road, grassland degradation, soil pollution and the like are caused.
The application also provides a grassland surface covering classification system based on transfer learning, and the grassland surface covering classification system comprises: the data acquisition module is used for acquiring a grassland remote sensing image of the target area; the preprocessing module is used for preprocessing the grassland remote sensing image to obtain preprocessed grassland remote sensing image data; the classification processing module comprises a trained grassland remote sensing image classification model and is configured to output image categories according to preprocessed grassland remote sensing image data; wherein the image category is one of preset ground surface covering categories. In an embodiment of the present application, the preset ground surface covering categories include: roads (including residential roads and illegal roads), grasslands, sandy roads, forests, and edges between grasslands, sandy roads, and forests. Wherein the grassland is green or dark green; the sand is white or light yellow, irregular in shape and has obvious sand ripple; the forest land is dark and speckled, and has irregular geometric shape and uneven texture; the road (including residential areas, illegal roads and the like) has larger values of brightness and length-width ratio, reflects the degrees and the variation trends of grassland degeneration, land desertification, sand prevention and control and soil pollution in the current research area through the variation analysis of various quantities per year, and laterally reflects the implementation effects of ecological related policies such as grassland compensation, seal protection, sand prevention and control engineering, seal withdrawal and the like in the research area.
Further, the preprocessing module is specifically configured to: carrying out true color band synthesis on remote sensing image data of the grassland through corresponding software according to the sequence of red, green and blue (RGB); deriving the image synthesized by the wave bands into a TIF format image according to a preset proportion (for example, 1: 100000); and cutting the derived TIF format image into a plurality of images with preset sizes (for example, 128-128) to obtain preprocessed remote-sensing grassland image data. For a specific embodiment of the grassland surface covering classification system, refer to the above description, and are not repeated herein.
The migration learning algorithm is mainly applied to carry out classification research on grassland earth surface coverings through high-resolution remote sensing images, classification results are analyzed, the range change condition of various earth surface coverings can be obtained, accurate, complete and rich scientific basis is provided for relevant departments, the utilization rate of grassland natural resources can be improved by implementing relevant policies, the effective implementation of relevant protection policies such as grassland ecological reward is guaranteed, reasonable utilization of grasslands is strengthened, ecological construction is strengthened, ecological rewarding is implemented, expected targets are achieved, and meanwhile the execution cost can be greatly reduced.
The above embodiments are only exemplary embodiments of the present application, and are not intended to limit the present application, and the protection scope of the present application is defined by the claims. Various modifications and equivalents may be made by those skilled in the art within the spirit and scope of the present application and such modifications and equivalents should also be considered to be within the scope of the present application.
Claims (10)
1. A grassland surface covering classification method based on transfer learning is characterized by comprising the following steps:
obtaining a grassland remote sensing image of a target area;
preprocessing the grassland remote sensing image to obtain preprocessed grassland remote sensing image data;
inputting the preprocessed grassland remote sensing image data into a trained grassland remote sensing image classification model, and outputting an image category;
wherein the image category is one of preset earth surface covering categories.
2. The method for classifying grassland surface coverings according to claim 1, wherein the step of preprocessing the grassland remote sensing image to obtain preprocessed grassland remote sensing image data comprises the following steps:
carrying out true color band synthesis on remote sensing image data of the grassland through corresponding software according to the sequence of red, green and blue (RGB);
deriving the images synthesized by the wave bands into TIF format images according to a preset proportion;
and cutting the derived TIF format image into a plurality of images with preset sizes to obtain preprocessed remote sensing image data of the grassland.
3. The grassland surface covering classification method according to claim 2, wherein the preset ratio is 1: 100000;
and/or the predetermined size is 128 x 128;
and/or the preset earth surface covering categories comprise: roads, grasses, sand, forests, and edges between grasses, sand, and forests.
4. The grassland surface covering classification method according to any one of claims 1 to 3, wherein the training method of the grassland remote sensing image classification model comprises the following steps:
constructing a training data set;
building a VGG network model based on transfer learning as a grassland remote sensing image classification model to be trained;
and inputting the training data set into the grassland remote sensing image classification model to be trained so as to train the grassland remote sensing image classification model.
5. The method of grassland surface covering classification of claim 4, wherein the constructing a training data set comprises:
acquiring remote sensing image data of a grassland;
carrying out true color band synthesis on remote sensing image data of the grassland through corresponding software according to the sequence of red, green and blue (RGB);
deriving the images synthesized by the wave bands into TIF format images according to a preset proportion;
cutting the derived TIF format image into a plurality of images with preset sizes;
screening the cut image, deleting the white image, and reserving the image containing the image information;
rotating the screened image for multiple times at different angles, and storing the rotated image for each time;
and marking categories of the screened image data according to the preset earth surface covering categories, and dividing the image data after the categories are marked into a training set, a verification set and a test set to obtain a training data set meeting the requirements.
6. The grassland surface covering classification method according to claim 5, wherein the building of a VGG network model based on transfer learning as the classification model of the remote sensing images of the grassland to be trained comprises the following steps:
reading a VGG network model which adopts a VGG19 model;
and connecting two full-connection layers behind the VGG19 to obtain the grassland remote sensing image classification model to be trained.
7. The method for classifying grassland surface coverings according to claim 6, wherein inputting a training data set into a grassland remote sensing image classification model to be trained to train the grassland remote sensing image classification model comprises:
training a grassland remote sensing image classification model to be trained by taking training set data as a training sample;
determining network structure and control parameters using the validation set data;
after the grassland remote sensing image classification model is trained, the test set data is input into the grassland remote sensing image classification model to test whether the output image category meets the requirements.
8. A grassland surface covering classification system based on transfer learning is characterized by comprising the following components:
the data acquisition module is used for acquiring a grassland remote sensing image of a target area;
the preprocessing module is used for preprocessing the grassland remote sensing image to obtain preprocessed grassland remote sensing image data;
the classification processing module comprises a trained grassland remote sensing image classification model and is configured to output an image category according to the preprocessed grassland remote sensing image data;
wherein the image category is one of preset earth surface covering categories.
9. The grassland surface covering classification system of claim 8, wherein the preprocessing module is specifically configured to:
carrying out true color band synthesis on remote sensing image data of the grassland through corresponding software according to the sequence of red, green and blue (RGB);
deriving the images synthesized by the wave bands into TIF format images according to a preset proportion;
and cutting the derived TIF format image into a plurality of images with preset sizes to obtain preprocessed remote sensing image data of the grassland.
10. The grassland surface covering classification system of claim 9, wherein the preset ratio is 1: 100000;
and/or the predetermined size is 128 x 128;
and/or the preset earth surface covering categories comprise: roads, grasses, sand, forests, and edges between grasses, sand, and forests.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911183095.4A CN110879992A (en) | 2019-11-27 | 2019-11-27 | Grassland surface covering object classification method and system based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911183095.4A CN110879992A (en) | 2019-11-27 | 2019-11-27 | Grassland surface covering object classification method and system based on transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110879992A true CN110879992A (en) | 2020-03-13 |
Family
ID=69730197
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911183095.4A Pending CN110879992A (en) | 2019-11-27 | 2019-11-27 | Grassland surface covering object classification method and system based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110879992A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111401275A (en) * | 2020-03-20 | 2020-07-10 | 内蒙古工业大学 | Information processing method and device for identifying grassland edge |
| CN112329852A (en) * | 2020-11-05 | 2021-02-05 | 西安泽塔云科技股份有限公司 | Classification method and device for earth surface coverage images and electronic equipment |
| CN113191302A (en) * | 2021-05-14 | 2021-07-30 | 成都鸿钰网络科技有限公司 | Method and system for monitoring grassland ecology |
| CN113361398A (en) * | 2021-06-04 | 2021-09-07 | 内蒙古工业大学 | Grassland fence identification method and device and storage medium |
| CN117312998A (en) * | 2023-09-06 | 2023-12-29 | 东北林业大学 | Road surface covering classification device and method based on image-weather-temperature |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108537192A (en) * | 2018-04-17 | 2018-09-14 | 福州大学 | A kind of remote sensing image ground mulching sorting technique based on full convolutional network |
| CN108875794A (en) * | 2018-05-25 | 2018-11-23 | 中国人民解放军国防科技大学 | Image visibility detection method based on transfer learning |
| CN109919206A (en) * | 2019-02-25 | 2019-06-21 | 武汉大学 | A kind of remote sensing image ground mulching classification method based on complete empty convolutional neural networks |
| CN110033045A (en) * | 2019-04-17 | 2019-07-19 | 内蒙古工业大学 | A kind of method and apparatus of trained identification image atomization |
-
2019
- 2019-11-27 CN CN201911183095.4A patent/CN110879992A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108537192A (en) * | 2018-04-17 | 2018-09-14 | 福州大学 | A kind of remote sensing image ground mulching sorting technique based on full convolutional network |
| CN108875794A (en) * | 2018-05-25 | 2018-11-23 | 中国人民解放军国防科技大学 | Image visibility detection method based on transfer learning |
| CN109919206A (en) * | 2019-02-25 | 2019-06-21 | 武汉大学 | A kind of remote sensing image ground mulching classification method based on complete empty convolutional neural networks |
| CN110033045A (en) * | 2019-04-17 | 2019-07-19 | 内蒙古工业大学 | A kind of method and apparatus of trained identification image atomization |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111401275A (en) * | 2020-03-20 | 2020-07-10 | 内蒙古工业大学 | Information processing method and device for identifying grassland edge |
| CN111401275B (en) * | 2020-03-20 | 2022-11-25 | 内蒙古工业大学 | Information processing method and device for identifying grassland edge |
| CN112329852A (en) * | 2020-11-05 | 2021-02-05 | 西安泽塔云科技股份有限公司 | Classification method and device for earth surface coverage images and electronic equipment |
| CN112329852B (en) * | 2020-11-05 | 2022-04-05 | 西安锐思数智科技股份有限公司 | Classification method and device for earth surface coverage images and electronic equipment |
| CN113191302A (en) * | 2021-05-14 | 2021-07-30 | 成都鸿钰网络科技有限公司 | Method and system for monitoring grassland ecology |
| CN113191302B (en) * | 2021-05-14 | 2022-11-01 | 成都鸿钰网络科技有限公司 | Method and system for monitoring grassland ecology |
| CN113361398A (en) * | 2021-06-04 | 2021-09-07 | 内蒙古工业大学 | Grassland fence identification method and device and storage medium |
| CN117312998A (en) * | 2023-09-06 | 2023-12-29 | 东北林业大学 | Road surface covering classification device and method based on image-weather-temperature |
| CN117312998B (en) * | 2023-09-06 | 2024-04-16 | 东北林业大学 | Road surface covering classification device and method based on image-weather-temperature |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110879992A (en) | Grassland surface covering object classification method and system based on transfer learning | |
| Ok et al. | Evaluation of random forest method for agricultural crop classification | |
| CN101840581B (en) | Method for extracting profile of building from satellite remote sensing image | |
| CN109829425B (en) | Farmland landscape small-scale ground feature classification method and system | |
| CN107832797A (en) | Classification of Multispectral Images method based on depth integration residual error net | |
| CN111563430A (en) | Public welfare forest land image intelligent identification method and system based on convolutional nerves | |
| Linhui et al. | Extracting the forest type from remote sensing images by random forest | |
| Entwisle et al. | Population growth and its spatial distribution as factors in the deforestation of Nang Rong, Thailand | |
| CN107194857B (en) | Available land resource analysis system based on high-resolution satellite remote sensing data | |
| CN109035223A (en) | A kind of intelligent evaluation method for satellite remote sensing images availability | |
| Bagan et al. | Improved subspace classification method for multispectral remote sensing image classification | |
| Qadri et al. | A comparative study of land cover classification by using multispectral and texture data | |
| CN112949738B (en) | Multi-class unbalanced hyperspectral image classification method based on EECNN algorithm | |
| CN112528726B (en) | Cotton aphid pest monitoring method and system based on spectral imaging and deep learning | |
| CN112927252A (en) | Newly-added construction land monitoring method and device | |
| CN116091850B (en) | Mining area land coverage classification model establishment and classification method | |
| Roy et al. | Comparative analysis of object based and pixel based classification for mapping of mango orchards in Sitapur district of Uttar Pradesh | |
| Mukhtar | Land use land cover change detection through remote sensing approach in Kano State Nigeria | |
| CN115115948A (en) | Forest land information fine extraction method based on random forest and auxiliary factors | |
| Hashim et al. | Multi-level image segmentation for urban land-cover classifications | |
| Gwata | Developing high resolution clutter for wireless network propagation using WorldView-2 imagery | |
| Bruce | Object oriented classification: case studies using different image types with different spatial resolutions | |
| Colaninno et al. | Defining Densities for Urban Residential Texture, through Land use Classification, from Landsat TM Imagery: Case Study of Spanish Mediterranean Coast | |
| Bueno et al. | Land use/land cover classification in a heterogeneous agricultural landscape using PlanetScope data. | |
| Kaplan | Semi-automatic multi-segmentation classification for land cover change dynamics in North Macedonia from 1988 to 2014 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200313 |