CN114255247A - Hilly land block depth segmentation and extraction method based on improved Unet + + network model - Google Patents
Hilly land block depth segmentation and extraction method based on improved Unet + + network model Download PDFInfo
- Publication number
- CN114255247A CN114255247A CN202111576669.1A CN202111576669A CN114255247A CN 114255247 A CN114255247 A CN 114255247A CN 202111576669 A CN202111576669 A CN 202111576669A CN 114255247 A CN114255247 A CN 114255247A
- Authority
- CN
- China
- Prior art keywords
- segmentation
- unet
- learning rate
- network model
- data
- 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
- 230000011218 segmentation Effects 0.000 title claims abstract description 39
- 238000000605 extraction Methods 0.000 title claims abstract description 21
- 238000000034 method Methods 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 27
- 238000000137 annealing Methods 0.000 claims abstract description 25
- 238000012795 verification Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000003709 image segmentation Methods 0.000 claims abstract description 5
- 230000000007 visual effect Effects 0.000 claims abstract description 5
- 238000011835 investigation Methods 0.000 claims abstract description 4
- 238000011160 research Methods 0.000 claims abstract description 4
- 230000000694 effects Effects 0.000 claims description 11
- 238000012937 correction Methods 0.000 claims description 5
- 230000003247 decreasing effect Effects 0.000 claims description 5
- 238000007689 inspection Methods 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims 1
- 238000012544 monitoring process Methods 0.000 description 6
- 235000002566 Capsicum Nutrition 0.000 description 4
- 239000006002 Pepper Substances 0.000 description 4
- 241000722363 Piper Species 0.000 description 4
- 235000016761 Piper aduncum Nutrition 0.000 description 4
- 235000017804 Piper guineense Nutrition 0.000 description 4
- 235000008184 Piper nigrum Nutrition 0.000 description 4
- 150000003839 salts Chemical class 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 238000012876 topography Methods 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000013467 fragmentation Methods 0.000 description 1
- 238000006062 fragmentation reaction Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000000399 orthopedic effect Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- 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
- 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
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
-
- 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
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
- G06T2207/20192—Edge enhancement; Edge preservation
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30188—Vegetation; Agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a hilly land block depth segmentation and extraction method based on an improved Unet + + network model, which comprises the following steps of: step 1: acquiring remote sensing image data of GF-1 in a research area and preprocessing the remote sensing image data; step 2: combining field investigation and visual interpretation, and making data sets with different segmentation scales by using a multi-scale segmentation method in image segmentation; and step 3: comparing the optimal segmentation scale, and correcting again to meet the precision requirement of the training set; and 4, step 4: cutting the corrected training set image by adopting a regular grid; and 5: and filling the data through a data enhancement operation, and performing data enhancement according to the following steps of 4: 1, dividing the image into a training set and a verification set; step 6: modeling by using a Unet + + network based on a cosine annealing learning rate; the method has the advantages of simple principle of establishing the Unet + + network model based on the cosine annealing learning rate, convenient operation, higher flexibility and practicability and stronger universality.
Description
Technical Field
The invention belongs to the technical field of a hilly land depth segmentation and extraction method, and particularly relates to a hilly land depth segmentation and extraction method of a Unet + + network model based on a cosine annealing learning rate.
Background
The cultivated land is an indispensable resource in human survival development, and the real-time and dynamic monitoring of cultivated land information is basic data of crop area monitoring, growth monitoring and yield prediction. With the rapid development and rapid acquisition of the satellite and the high-resolution image, the possibility is provided for accurately and effectively monitoring farmland information in a large range. In the 70 s of the 20 th century, expert and scholars in the european and american countries have begun to study the extraction of farmland information using remote sensing data. However, traditional remote sensing information extraction, including maximum likelihood, support vector machine, random forest and the like, is based on pixel classification, which causes salt and pepper phenomena, and the integrity of the land, the boundary information and the extraction precision thereof are low. Although the visual interpretation method can obtain higher accuracy of the plot information, a large amount of manpower and material resources are consumed, the information obtaining period is long, and the real-time and fine requirements of land monitoring, crop disaster damage assessment and the like are difficult to meet. In the existing research, farmland information extraction has obtained good classification results in densely planted and single plain areas, however, cultivated land characteristics of China are complex and various, especially terrain relief of hilly areas is large, land block breaking degree is high, additionally, planting area is different from land block boundary shape, terrain conditions are complex, so that mountain farmland information is difficult to obtain quickly and accurately, and the quick automatic extraction of mountain farmland information based on a traditional remote sensing monitoring method is difficult, which brings great challenges to the accurate extraction of remote sensing cultivated land information. Meanwhile, due to the fact that the hilly area topography fluctuates and forms shadows on the image, the spectral characteristics of the slope farmland are very complex, and the slope farmland presents different colors on the image. If the training samples of the slope farmland are manually selected by adopting conventional supervision and classification, the workload is large, and the samples of the slope farmland are difficult to accurately obtain, so that the classification precision is influenced. Although the object-oriented classification method fully utilizes the relevant characteristics of the image object to complete the multi-scale segmentation of the image, the salt and pepper phenomenon in the classification process can be effectively reduced, and the ground feature classification precision is improved, the image segmentation scale under different resolution images and planting conditions is often difficult to accurately determine, and a better segmentation effect is often difficult to obtain in an area with high degree of fragmentation.
Disclosure of Invention
In order to solve the defects in the technical problems, the invention provides a method for deeply segmenting and extracting the hilly land parcel based on a Unet + + network model of cosine annealing learning rate, which can improve the precision of deeply segmenting and extracting the land parcel to a certain extent.
In order to solve the technical problems, the invention adopts the technical scheme that:
a hilly land block depth segmentation and extraction method based on an improved Unet + + network model comprises the following steps:
step 1: acquiring high-resolution first-grade (GF-1) remote sensing image data in a research area and preprocessing the data; the data preprocessing comprises the following steps: radiometric calibration, atmospheric correction, orthometric correction, image fusion, clipping, etc.;
step 2: combining field investigation and visual interpretation, utilizing a multi-scale segmentation method in image segmentation, and combining parameters such as segmentation scale, shape factor and the like to respectively make segmentation scales of 40, 60 and 80;
and step 3: comparing the optimal segmentation scale, and if the situation of mistaken segmentation of field roads, greenhouses and the like still exists; after manual inspection and editing, performing ground verification again, and correcting again to enable the ground verification to meet the precision requirement of the training set;
and 4, step 4: cutting the corrected training set image into 256 multiplied by 256 images and constructing a label training data set and training a model by adopting a regular grid;
and 5: data are filled through data enhancement operations such as horizontal turning, vertical turning, diagonal images and the like, and the data are filled according to the following steps of 4: 1, dividing the image into a training set and a verification set;
step 6: establishing a Unet + + network model based on a cosine annealing learning rate, and replacing the learning rate of a fixed constant with the cosine annealing learning rate; the cosine annealing learning rate is firstly decreased at the highest speed along with the increase of the iteration times, then is increased suddenly, and then is continuously repeated, so that the purpose is to escape from a local optimum point, the convergence rate of the model is favorably accelerated, the model effect is better, and the formula is as follows:
where i is the index value of the run, ηmini and ηmaxi is the minimum value and the maximum value of the learning rate respectively, and defines the range of the learning rate; tcur indicates how many epochs are currently executed, and is updated after each batch run; ti indicates the total epoch number in the ith run.
In the step 4, the corrected training set image is cut by a regular grid method to form 256 × 256 images and label training data sets, and model training is performed.
According to the method for deeply dividing and extracting the hilly land parcel based on the improved Unet + + network model, in the step 3, after manual inspection and editing, ground verification is carried out again, 325 sample points are investigated in total, 16 errors exist, the field road and the greenhouse are divided into farmland by mistake, and the total precision of the method reaches 95.07%.
In the method for deep segmentation and extraction of hilly land blocks based on the improved Unet + + network model, in the step 6, a Unet + + network based on a cosine annealing learning rate is used for modeling. The Unet + + designs a series of nested dense convolution fast and hopping connections, changes the connectivity of encoder and decoder networks, and introduces a deep supervision scheme to supervise the output of each Unet + + branch. The invention replaces the learning rate of the fixed constant with the cosine annealing learning rate. The cosine annealing learning rate is decreased at the highest speed along with the increase of the iteration times, then is increased suddenly, and then is repeated continuously, so that the purpose is to escape from a local optimum point, the convergence rate of the model is increased, and the model effect is better;
according to the method for deeply segmenting and extracting the hilly land parcel based on the improved Unet + + network model, a Kappa coefficient, Overall Accuracy (OA), User Accuracy (UA), Producer Accuracy (PA) and F1 scores are selected as accuracy evaluation indexes of classification results, and the classification capability of the Unet + + network model based on the cosine annealing learning rate is verified, so that the purpose of deeply segmenting and extracting the land parcel with high accuracy is achieved.
Kappa coefficient:
po is the sum of the number of correctly classified samples of each class divided by the total number of samples, i.e. the overall classification accuracy. The number of real samples of each class is a1, a 2.,. aC, and the number of samples of each class predicted is b1, b 2.,. bC; n is the total number of samples.
OA (overall accuracy):
true Positive (TP): predicting the positive class as a positive class number;
true Negative, TN: predicting a negative class as a negative class number;
false Positive (FP): predicting the negative class as a positive class number false alarm (Type I error);
false Negative (FN): predict positive class as negative class number → false negative (Type II error).
User precision:
the precision of a producer:
f1 score:
the invention has the following beneficial effects: the difficult point of land parcel segmentation in hilly and mountainous areas lies in: 1. the terrain is complex, the land parcel is broken, and the planting area is different from the boundary shape of the land parcel; 2. the spectral characteristics of the slope land are very complicated due to the fact that the hilly area topography fluctuates to form shadows on the image. The Unet + + network based on the cosine annealing learning rate can solve the difficult point of land block segmentation in hilly and mountainous areas, better maintain the smoothness and integrity of the boundary of the hilly land blocks and improve the classification precision. The method utilizes the cosine annealing learning rate to replace the learning rate of a fixed constant, combines the Unet + + network deep learning model, improves the precision of deep segmentation and extraction of hilly plots, has strong universality and is easier to popularize and apply.
Drawings
FIG. 1 shows the classification results of hilly lands with the division scale of 40, 60 and 80;
FIG. 2 is a curve of the learning rate variation of cosine annealing;
FIG. 3 is a Unet + + network model architecture;
FIG. 4 shows the results of the true annotation and 3 classification methods
Detailed Description
The present invention will be described in detail with reference to specific examples.
Step 1: acquiring high-grade first-grade (GF-1) remote sensing image data of Qingshui river county and Yangyuan county. The high-score first image data of 2019 and 2021 years were downloaded from the terrestrial observation satellite center website using the high-score first image of L1A level. And preprocessing operations such as Radiometric Calibration, Atmospheric Calibration, Orthorectification, image fusion, clipping and the like are carried out in ENVI by utilizing Radiometric Calibration, FLAASH Atmospheric Correction, RPC orthopedic Calibration Workflow and NNDiffuse Pan shading modules.
Step 2: in conjunction with the field investigation and visual interpretation, data sets of different segmentation scales were created by a multi-scale segmentation method in image segmentation, and segmentation scales of 40, 60, and 80 were created, respectively, as shown in fig. 1.
And step 3: when the segmentation scale is 60, the integrity of the map spot is maintained (the upper right yellow square), and ground objects such as non-crop areas (the lower left light blue square) in the farmland and the like of the road are effectively extracted, so that the segmentation effect is the best (as shown in fig. 1). The optimal segmentation scale is compared, but the situation that the field road, the greenhouse and the like are wrongly divided still exists. After manual inspection and editing, ground verification is carried out again, 325 sample points are investigated totally, 16 errors are included, the field road and the greenhouse are divided into farmland by mistake, and the total precision of the method reaches 95.07%.
And 4, step 4: and constructing a 256 multiplied by 256 size image and label training data set and training a model by adopting a regular grid cutting method for the corrected training set image.
And 5: data are filled through data enhancement operations such as horizontal turning, vertical turning, diagonal images and the like, and the data are filled according to the following steps of 4: a scale of 1 divides the images into training and validation sets with sample numbers of 3520 and 880 images, respectively.
Step 6: and the method optimizes the Unet + + network model by using the cosine annealing learning rate and improves the segmentation precision of the Unet + + network. The cosine annealing learning rate is adopted, along with the increase of iteration times, the cosine annealing learning rate is decreased at the highest speed, then suddenly increased, and then the process is continuously repeated, so that the purpose is to escape from a local optimum point, the convergence rate of the model is favorably accelerated, the model effect is better (as shown in fig. 2), and the formula is as follows:
where i is the index value of the run, ηmini and ηmaxi is the minimum value and the maximum value of the learning rate respectively, and defines the range of the learning rate; tcur indicates how many epochs are currently executed, and is updated after each batch run; ti indicates the total epoch number in the ith run.
Unet + + (fig. 3) redesigns a series of nested dense volume blocks and hopping connections, changes the connectivity of the encoder and decoder networks, and introduces a deep supervision scheme to supervise the output of each Unet + + branch.
The Kappa coefficient, the Overall Accuracy (OA), the User Accuracy (UA), the Producer Accuracy (PA) and the F1 score are used as the accuracy evaluation indexes of the classification result. The formula is as follows:
kappa coefficient:
po is the sum of the number of correctly classified samples of each class divided by the total number of samples, i.e. the overall classification accuracy. The number of real samples of each class is a1, a 2.,. aC, and the number of samples of each class predicted is b1, b 2.,. bC; n is the total number of samples.
OA (overall accuracy):
true Positive (TP): predicting the positive class as a positive class number;
true Negative, TN: predicting a negative class as a negative class number;
false Positive (FP): predicting the negative class as a positive class number false alarm (Type I error);
false Negative (FN): predict positive class as negative class number → false negative (Type II error).
User precision:
the precision of a producer:
f1 score:
as can be seen from fig. 4 and table 1, the conventional SVM has a poor classification effect, and shows more salt and pepper phenomena, and the classification effect shows more discontinuous small blocks.
TABLE 1
The deep learning model is better in the definition of the edges of the land parcels, the integrity of the land parcels and the classification precision. The traditional fixed constant learning rate is gradually reduced in the training process, a local optimal point is gradually found, and a better local optimal point is converged. Compared with the unet + + model using a learning rate of a fixed constant, the unet + + model using the cosine annealing learning rate quickly steps at a local optimum point first, and the model is saved. After the model is saved, the learning rate is restored to a large value again, a new optimal point is searched, and due to the fact that the models of the optimal points of different local models have large diversity, the effect is better after the models are integrated, the overall effect of the models can be improved, and redundant training cost cannot be generated. Compared with a unet + + model with a fixed constant learning rate, the overall accuracy of the improved unet + + is improved by 3.6%, and the F1 score is improved by 3.29%. The overall accuracy of the improved unet + + network reaches 92.75%, which is 7.6% and 2.5% higher than that of the SVM and unet models, respectively, the F1 score is 10.26% and 2.01% higher than that of the SVM and unet models, respectively, and the Kappa coefficient is 20% and 2% higher than that of the SVM and unet models, respectively. In the aspect of operation speed, the SVM can quickly fit a proper amount of sample data, but the prediction speed is low, the classification precision is low, and the salt and pepper phenomenon is easy to occur. Deep learning is more efficient in predicting, although the training model is longer.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.
Claims (6)
1. A hilly land block depth segmentation and extraction method based on an improved Unet + + network model is characterized by comprising the following steps:
step 1: acquiring high-resolution first-grade (GF-1) remote sensing image data in a research area and preprocessing the data; the data preprocessing comprises the following steps: radiometric calibration, atmospheric correction, orthometric correction, image fusion, clipping, etc.;
step 2: combining field investigation and visual interpretation, utilizing a multi-scale segmentation method in image segmentation, and combining segmentation scales and shape factor parameters to respectively perform different segmentation scales;
and step 3: comparing the optimal segmentation scale, and if the situation of mistaken segmentation of field roads, greenhouses and the like still exists; after manual inspection and editing, performing ground verification again, and correcting again to enable the ground verification to meet the precision requirement of the training set;
and 4, step 4: cutting the corrected training set image into 256 multiplied by 256 images and constructing a label training data set and training a model by adopting a regular grid;
and 5: data are filled through data enhancement operations such as horizontal turning, vertical turning, diagonal images and the like, and the data are filled according to the following steps of 4: 1, dividing the image into a training set and a verification set;
step 6: establishing a Unet + + network model based on a cosine annealing learning rate, and replacing the learning rate of a fixed constant with the cosine annealing learning rate; the cosine annealing learning rate is firstly decreased at the highest speed along with the increase of the iteration times, then is increased suddenly, and then is continuously repeated, so that the purpose is to escape from a local optimum point, the convergence rate of the model is favorably accelerated, the model effect is better, and the formula is as follows:
where i is the index value of the number of runs,
and
respectively defining the range of the learning rate for the minimum value and the maximum value of the learning rate; tcur indicates how many epochs are currently executed, and is updated after each batch run; ti indicates the total epoch number in the ith run.
2. The method for deep segmentation and extraction of hilly land parcel based on improved Unet + + network model as claimed in claim 1, wherein in step 4, the modified training set image is cut by regular grid to form 256 x 256 size images and label training data set for construction and model training.
3. The method for deep segmentation and extraction of hilly land parcel based on the improved Unet + + network model as claimed in claim 1, wherein in step 3, after manual inspection and editing, ground verification is performed again, and 325 sample points are investigated in total, wherein 16 of them are mistakenly divided into cultivated land for field road and cultivated land for greenhouse, respectively, and the total precision reaches 95.07%.
4. The method for deep segmentation and extraction of hilly areas based on the improved Unet + + network model as claimed in claim 1, wherein in step 6, the Unet + + network based on the cosine-annealing learning rate is used for modeling. A series of nested dense convolution fast and jump connections are designed by the Unet + +, the connectivity of an encoder and a decoder network is changed, and a deep supervision scheme is introduced to supervise the output of each Unet + + branch; the invention uses the cosine annealing learning rate to replace the learning rate of the fixed constant; the cosine annealing learning rate is decreased at the highest speed along with the increase of the iteration times, then is increased suddenly, and then is repeated continuously, so that the purpose is to escape from a local optimum point, the convergence rate of the model is increased, and the model effect is better.
5. The method for deep segmentation and extraction of hilly terrain based on the improved Unet + + network model as claimed in claim 1, wherein the Kappa coefficient, Overall Accuracy (OA), User Accuracy (UA), Producer Accuracy (PA) and F1 scores are selected as accuracy evaluation indexes of classification results, and the classification capability of the Unet + + network model based on the cosine annealing learning rate is verified to achieve the purpose of deep segmentation and extraction of high-accuracy terrain.
6. The method for deep segmentation and extraction of hilly terrain blocks based on the improved Unet + + network model as claimed in claim 5,
kappa coefficient:
po is the sum of the number of correctly classified samples of each class divided by the total number of samples, i.e. the overall classification accuracy. The number of real samples of each class is a1, a 2.,. aC, and the number of samples of each class predicted is b1, b 2.,. bC; n is the total number of samples;
OA (overall accuracy):
true Positive (TP): predicting the positive class as a positive class number;
true Negative, TN: predicting a negative class as a negative class number;
false Positive (FP): predicting the negative class as a positive class number false alarm (Type I error);
false Negative (FN): predict positive class as negative class number → false negative (Type II error).
User precision:
the precision of a producer:
f1 score:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111576669.1A CN114255247A (en) | 2021-12-22 | 2021-12-22 | Hilly land block depth segmentation and extraction method based on improved Unet + + network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111576669.1A CN114255247A (en) | 2021-12-22 | 2021-12-22 | Hilly land block depth segmentation and extraction method based on improved Unet + + network model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114255247A true CN114255247A (en) | 2022-03-29 |
Family
ID=80794019
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111576669.1A Pending CN114255247A (en) | 2021-12-22 | 2021-12-22 | Hilly land block depth segmentation and extraction method based on improved Unet + + network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114255247A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114882306A (en) * | 2022-04-06 | 2022-08-09 | 国家基础地理信息中心 | Topographic map scale identification method and device, storage medium and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111275020A (en) * | 2020-03-12 | 2020-06-12 | 青梧桐有限责任公司 | Room state identification method |
| CN111723711A (en) * | 2020-06-10 | 2020-09-29 | 内蒙古农业大学 | Plianes and object-oriented mulching film information extraction method and system |
| CN113204608A (en) * | 2021-05-27 | 2021-08-03 | 广州大学 | Automatic map updating method, storage medium and system based on remote sensing image |
| CN113516084A (en) * | 2021-07-20 | 2021-10-19 | 海南长光卫星信息技术有限公司 | High-resolution remote sensing image semi-supervised classification method, device, equipment and medium |
-
2021
- 2021-12-22 CN CN202111576669.1A patent/CN114255247A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111275020A (en) * | 2020-03-12 | 2020-06-12 | 青梧桐有限责任公司 | Room state identification method |
| CN111723711A (en) * | 2020-06-10 | 2020-09-29 | 内蒙古农业大学 | Plianes and object-oriented mulching film information extraction method and system |
| CN113204608A (en) * | 2021-05-27 | 2021-08-03 | 广州大学 | Automatic map updating method, storage medium and system based on remote sensing image |
| CN113516084A (en) * | 2021-07-20 | 2021-10-19 | 海南长光卫星信息技术有限公司 | High-resolution remote sensing image semi-supervised classification method, device, equipment and medium |
Non-Patent Citations (4)
| Title |
|---|
| ZONGWEI ZHOU等: ""UNet++: A Nested U-Net Architecture for Medical Image Segmentation"", 《ARXIV》 * |
| 张强等: "《基于深度神经网络技术的高分遥感图像处理及应用》", 31 August 2020, 中国宇航出版社 * |
| 李勇等: "《复杂情感分析方法及其应用》", 30 April 2020, 冶金工业出版社 * |
| 焦李成等: "《人工智能、类脑计算与图像解译前沿》", 31 January 2020, 西安电子科技大学出版社 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114882306A (en) * | 2022-04-06 | 2022-08-09 | 国家基础地理信息中心 | Topographic map scale identification method and device, storage medium and electronic equipment |
| CN114882306B (en) * | 2022-04-06 | 2023-08-18 | 国家基础地理信息中心 | Topography scale identification method and device, storage medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108921885B (en) | Method for jointly inverting forest aboveground biomass by integrating three types of data sources | |
| CN106815847B (en) | Trees dividing method and single tree extracting method based on laser radar point cloud | |
| CN106815850A (en) | The method that canopy density forest reserves very high is obtained based on laser radar technique | |
| CN105740759B (en) | Semilate rice information decision tree classification approach based on feature extraction in multi-temporal data | |
| CN111598045B (en) | Remote sensing farmland change detection method based on object spectrum and mixed spectrum | |
| CN112749627A (en) | Method and device for dynamically monitoring tobacco based on multi-source remote sensing image | |
| CN107247927B (en) | Method and system for extracting coastline information of remote sensing image based on tassel cap transformation | |
| CN105894044A (en) | Single-plant tree point cloud automatic extraction method based on vehicle-mounted laser scanning data | |
| CN103034858A (en) | Secondary clustering segmentation method for satellite cloud picture | |
| CN104502919A (en) | Method for utilizing airborne laser radar point cloud to extract urban vegetation three-dimensional coverage map | |
| Jiang et al. | Estimation of land production and its response to cultivated land conversion in North China Plain | |
| CN105447274A (en) | Method of performing coastal wetland drawing for medium-resolution remote sensing image by utilizing object-oriented classification technology | |
| CN116129262A (en) | Cultivated land suitability evaluation method and system for suitable mechanized transformation | |
| Hu et al. | Integrating CART algorithm and multi-source remote sensing data to estimate sub-pixel impervious surface coverage: a case study from Beijing Municipality, China | |
| CN111222536A (en) | City green space information extraction method based on decision tree classification | |
| CN110569805B (en) | Artificial forest stand wood extraction and standing quality evaluation method based on unmanned aerial vehicle image point cloud | |
| CN115496939A (en) | Wetland information extraction method based on multi-source remote sensing data | |
| CN114255247A (en) | Hilly land block depth segmentation and extraction method based on improved Unet + + network model | |
| Sun et al. | Feasibility study on the estimation of the living vegetation volume of individual street trees using terrestrial laser scanning | |
| Huang et al. | Recognition and counting of pitaya trees in karst mountain environment based on unmanned aerial vehicle RGB images | |
| CN110245694A (en) | A kind of the Northwest's agricultural film information analysis method based on high score Remote Sensing Image Segmentation | |
| CN115115948B (en) | Forest land information refined extraction method based on random forest and auxiliary factors | |
| CN107403057A (en) | A kind of Chinese medicine Quality Regionalization model based on maximum informational entropy and improved independence weight coefficient | |
| Sun et al. | TPRoGI: a comprehensive rock glacier inventory for the Tibetan Plateau using deep learning | |
| Trier et al. | Semi-automatic detection of burial mounds in forested areas |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220329 |