CN110503020A - A method of based on deep learning air remote sensing target detection - Google Patents
A method of based on deep learning air remote sensing target detection Download PDFInfo
- Publication number
- CN110503020A CN110503020A CN201910757510.6A CN201910757510A CN110503020A CN 110503020 A CN110503020 A CN 110503020A CN 201910757510 A CN201910757510 A CN 201910757510A CN 110503020 A CN110503020 A CN 110503020A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sample
- neural network
- deep learning
- target detection
- 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
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000001514 detection method Methods 0.000 title claims abstract description 33
- 238000013135 deep learning Methods 0.000 title claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 17
- 238000012360 testing method Methods 0.000 claims abstract description 13
- 238000005070 sampling Methods 0.000 claims abstract description 10
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims abstract description 8
- 238000010606 normalization Methods 0.000 claims abstract description 7
- 238000013434 data augmentation Methods 0.000 claims abstract description 6
- 230000009466 transformation Effects 0.000 claims abstract description 6
- 230000000717 retained effect Effects 0.000 claims abstract description 4
- 238000013461 design Methods 0.000 claims abstract description 3
- 238000013528 artificial neural network Methods 0.000 claims description 12
- 238000003062 neural network model Methods 0.000 claims description 8
- 230000003416 augmentation Effects 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 230000001629 suppression Effects 0.000 claims description 3
- 230000007306 turnover Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 2
- 238000011426 transformation method Methods 0.000 claims description 2
- 239000000284 extract Substances 0.000 abstract description 2
- 230000001737 promoting effect Effects 0.000 abstract 1
- 230000006870 function Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 241000209094 Oryza Species 0.000 description 2
- 235000007164 Oryza sativa Nutrition 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 235000009566 rice Nutrition 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000002352 surface water Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- 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
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/02—Affine transformations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4023—Scaling of whole images or parts thereof, e.g. expanding or contracting based on decimating pixels or lines of pixels; based on inserting pixels or lines of pixels
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/60—Rotation of whole images or parts thereof
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of methods based on deep learning air remote sensing target detection, including data augmentation, and to promote detection accuracy, wherein data augmentation uses affine transformation and field interpolation.For the test problems of Issues On Multi-scales especially Small object object, from the beginning training detection network, in a network using batch normalization, while promoting learning rate, can preferably adapt to Detection task in this way.Meanwhile design feature extracts network structure, in order to reduce the loss of local message, maximum pond is removed, carries out down-sampling with convolution replacement, while updating the convolution kernel in network structure, so that local message is retained, to improve detector to the detection performance of air remote sensing target.
Description
Technical field
The present invention relates to deep learning, image procossing, remote sensing fields, in particular for the Small object in aerial remote sensing images
The method for carrying out target detection.
Background technique
Currently, the quality and quantity of remote sensing images has very big promotion with the rapid development of remote sensing satellite technology, this
The data of a little substantial amounts become a kind of highly important resource, have a wide range of applications in civil and military field
With important researching value.Deep learning has become the important method of air remote sensing area research, receives domestic and international
Art circle and industry are more and more paid close attention to, and the target detection of remote sensing fields has huge application and researching value.
Remote sensing images have certain particularity, are scale diversity first, shooting height variation range from several hundred rice to
Nearly myriametre has, also, the size disparity of same category object is also very big, such as the ship on the water surface, argosy can only achieve
Several hundred rice, small boat often only have tens meters.Secondly, the surface water that the visual angle of remote sensing images is different from normal image is looked squarely
Angle is to take a crane shot from high-altitude mostly, so performance detects network well on routine data collection, detects air remote sensing picture
Effect not necessarily can be fine.There exists a further problem in that Small object problem, many objects only have tens very remote sensing images on the image
To several pixels, in addition, the operation of pond layer can also allow Small object to further reduce.The shooting angle of remote sensing images causes
The direction of target object is uncertain.On the other hand, the background complexity for including in the visual field of remote sensing images is high, therefore can be right
The detection of object causes bigger interference.These problems increase the difficulty of Remote Sensing Target detection.
Summary of the invention
The object of the present invention is to provide a kind of method based on deep learning air remote sensing target detection, the method reduce
Parameter amount, and detection speed is accelerated, so that time and computing redundancy degree are smaller, while Detection accuracy is also obtained and is mentioned
It rises.
Realizing the specific technical solution of the object of the invention is:
A method of based on deep learning air remote sensing target detection, this method comprising the following specific steps
Step 1: building Airborne Data Classification collection
Aerial remote sensing images are classified and marked, forms image and the one-to-one set of mark, it is distant to constitute aviation
Feel data set；Data augmentation is carried out to Airborne Data Classification collection using affine transformation and neighbor interpolation；Then by the number after augmentation
Test set and training set are divided into according to collection；
Step 2: building deep neural network structure
For the characteristic of remote sensing images, design construction deep neural network, and it is trained；
Step 3: according to constructed deep learning neural network, being tested using the Airborne Data Classification collection.
Data increasing is carried out to Airborne Data Classification collection using affine transformation and neighbor interpolation method described in step 1 of the present invention
Extensively, specifically:
Using affine transformation method, including cuts, rotates and turn over；
Using neighbor interpolation method, (xi, yi) and (xj,yj) it is the sample pair that initial data is concentrated, it is adopted at random from data
Sample obtains, and x is image pattern, and y is its corresponding label, and λ belongs to [0,1], It is newly-generated
Sample pair.
Deep neural network is constructed described in step 2 of the present invention, specifically:
Network structure based on ResNet, first first down-sampling operation of cancellation ResNet, eliminates maximum pond, uses
Convolution operation carries out down-sampling, replaces the convolution kernel in ResNet, so that wisp information is retained, to improve detector
To the detection performance of the target of small and dense collection distribution；Secondly, not using pre-training model, batch normalization is added in a network, makes
It is 0 that data, which become mean value, the distribution that standard deviation is 1；Also, due to joined batch normalization, learning rate, such energy can be promoted
It is enough to promote detection effect higher；Deep neural network model is trained.
Test process described in step 3 of the present invention includes:
Using the deep neural network constructed in the Airborne Data Classification collection and step 2 built in step 1, carries out end and arrive
The test at end:
1) picture to be detected is inputted, picture is input in sorter network to obtain various sizes of feature and reflect
It penetrates；
2) wherein six layers of characteristic pattern is extracted, each point respectively on these characteristic patterns constructs 6 different rulers later
The encirclement frame for spending size, is then detected and is classified respectively, generate multiple encirclement frames；
3) different characteristic figure encirclement frame obtained removed with non-maxima suppression method a part overlapping or
Incorrect encirclement frame generates final encirclement frame set；
4) result is evaluated and tested, FN: is judged as negative sample, but be in fact positive sample；FP: it is determined the sample that is positive
This, but be in fact negative sample；TN: being judged as negative sample, in fact and negative sample；TP: being judged as positive sample, true
Upper is also card sample；It calculates accuracy rate=TP/ (TP+FP), calculates recall rate=TP/ (TP+FN).When P is very high, R can be obvious
It reduces, on the contrary, P can be substantially reduced when R is very high, so using F1score come overall merit, F1score=2*P*R/ (P+
R)。
Compared to existing method, the beneficial effects of the present invention are:
1) the data augmentation method for the field interpolation that this method proposes, greatly improves generalization ability, reduces
The expense of unnecessary error label.
1) deep neural network model that this method proposes can extract original image loss partial information, especially wisp
Information defect, improve classification capacity by a relatively large margin.
2) deep neural network model that this method proposes, reduces parameter amount, and accelerate detection speed, so that when
Between and computing redundancy degree it is smaller, while Detection accuracy also obtains promotion.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Specific embodiment
The present invention will be further described with reference to the accompanying drawing.
Refering to fig. 1, the present invention the following steps are included:
1, Airborne Data Classification collection is constructed
Aerial remote sensing images are classified and marked, image and the one-to-one set of mark are generated.Become using radiation
Change method, the method that superposition Gaussian noise method and neighbor interpolation etc. newly propose carries out augmentation to the Airborne Data Classification collection.
Then data set is divided into training set and test set in proportion, is trained and tests in step below.
For the small problem of mark sample set quantity, existing data set is increased using affine transformation and neighbor interpolation
Extensively.
Wherein, neighbor interpolation is a kind of effective ways of the data set augmentation proposed, and detailed process is as follows:
In supervised learning, a function f can be gone under normal conditions to describe the feature vector, X needed and mesh
The relationship between vector Y is marked, that this relationship follows is Joint Distribution P (X, Y).It is therefore desirable to the loss defined using one
Function l (f (x), y) is come the difference punished between predicted value f (x) and real goal value y.Next, minimizing in this joint
The average loss being distributed on P, here it is expected risk R (f):
R (f)=∫ l (f (x), y) dP (x, y)
Although this distribution P be all in most cases it is ignorant, by using training set, can pass through through
Distributed acquisition is tested to approximate distribution Pδ(x, y):
It is distributed by using experience, the approximate expected risk R of empiric risk can be obtainedδ(f):
The function f that approximate expected risk by minimizing formula above learns is exactly traditional empirical risk minimization
Change theoretical.
For approximate true distribution there are many kinds of method, simplicity estimation is one of.And in vicinal risk minimization original
In then, distribution P is defined as follows:
Wherein v is a neighborhood distribution, for indicating imaginary to searching out on (xi, yi) neighborhood in training characteristics-target
Feature-target pair.
It joined a kind of new neighborhood distribution, it may be assumed that
In this case, resulting virtual feature-object vector can be expressed as follows:
In above-mentioned formula, (xi, yi) and (xj, yj) is the sample pair that initial data is concentrated, from from training data
Stochastical sampling, λ belong to [0,1].The effect of the hyper parameter λ of neighbor interpolation is degree of the controlling feature-target to interpolation.
2, training neural network target detection model:
Projected depth neural network model is trained using the Airborne Data Classification collection, obtains deep learning mind
Through network model.
The network structure based on ResNet is designed, due to the convolution kernel and maximum in ResNet structure, used at the beginning
Pondization operation carries out down-sampling, and original image can be reduced four times by such operation, so that the local message damage needed in the detection
It has lost very much.In order to reduce the loss of local message, the stride of first convolutional layer of ResNet is changed to 1 from 2 first,
Exactly cancel first down-sampling operation, eliminates maximum pond, carry out down-sampling with convolution operation, while replacing ResNet
In 7*7 convolution be 3 3x3 convolutional layers so that local message is retained, to improve detector to the mesh of small and dense collection distribution
Target detection performance.
During using batch normalization from the beginning training detection network, since the requirement of Detection task and classification task is not
With, firstly, abandoning the pre-training model of classification, detection model is trained from the beginning.Secondly, batch normalization is added, it is placed on sharp
Before function Relu living, one layer of output data x is sought1To xmMean μβ, seek the standard deviation of one layer of output dataM is the scale of this collection of training sample batch, is normalized, finds out normalization resultWherein ε be a very little close to 0 number, in order to prevent denominator be equal to zero.Also, learning rate is promoted,
It solves the above problems, can preferably adapt to Detection task in this way.
3, target detection is carried out using the neural network model that training obtains:
According to the deep learning neural network model, target detection is carried out to aerial remote sensing images.
A picture to be detected is inputted, picture is input in sorter network to obtain various sizes of Feature Mapping.
Wherein six layers of characteristic pattern is extracted, each point respectively on these characteristic patterns constructs six different scales later
The encirclement frame of size, is then detected and is classified respectively, and multiple encirclement frames are generated.
Different characteristic figure encirclement frame obtained is combined, a part weight is removed with non-maxima suppression method
Folded or incorrect encirclement frame, generates final encirclement frame set.
Result is evaluated and tested, FN: being judged as negative sample, but is in fact positive sample.FP: being judged as positive sample,
It but is in fact negative sample.TN: being judged as negative sample, in fact and negative sample.TP: it is judged as positive sample, in fact
It is also card sample.It calculates accuracy rate=TP/ (TP+FP), calculates recall rate=TP/ (TP+FN).When P is very high, R can obviously drop
It is low, on the contrary, P can be substantially reduced when R is very high, so using F1score come overall merit, F1score=2*P*R/ (P+R).
Embodiment
The present embodiment uses the 1200 picture samples from Google Earth, has expanded 6000 using cutting out, has made
7200 picture samples have been expanded with rotating and turn over, have expanded 7200 picture samples using neighbor interpolation, these picture samples
This constitutes the data set comprising 21600 picture samples altogether, and data set is divided into training set and test set, training set ratio
Example be 75%, i.e., 16200, test set ratio be 25%, i.e., 5400.
Followed by training set training deep neural network, obtain deep neural network model, then using test set into
Row test, is calculated multiple groups accuracy rate and recall rate, calculates F1score and reaches 0.89 ± 0.08.
Claims (4)
1. a kind of method based on deep learning air remote sensing target detection, which is characterized in that this method includes walking in detail below
It is rapid:
Step 1: building Airborne Data Classification collection
Aerial remote sensing images are classified and marked, forms image and the one-to-one set of mark, constitutes air remote sensing number
According to collection；Data augmentation is carried out to Airborne Data Classification collection using affine transformation and neighbor interpolation method；Then by the number after augmentation
Test set and training set are divided into according to collection；
Step 2: building deep neural network structure
For the characteristic of remote sensing images, design construction deep neural network, and it is trained；
Step 3: according to constructed deep learning neural network, being tested using the Airborne Data Classification collection.
2. the method for deep learning air remote sensing target detection according to claim 1, which is characterized in that described in step 1
Data augmentation is carried out to Airborne Data Classification collection using affine transformation and neighbor interpolation method, specifically:
Using affine transformation method, including cuts, rotates and turn over；
Using neighbor interpolation method, (xi, yi) and (xj,yj) it is the sample pair that initial data is concentrated, stochastical sampling obtains from data
It arriving, x is image pattern, and y is its corresponding label, and λ belongs to [0,1], It is newly-generated sample
It is right.
3. a kind of method based on deep learning air remote sensing target detection according to claim 1, which is characterized in that step
The rapid 2 building deep neural network, specifically:
Network structure based on ResNet, first first down-sampling operation of cancellation ResNet, eliminates maximum pond, uses convolution
It operates to carry out down-sampling, the convolution kernel in ResNet is replaced, so that wisp information is retained, to improve detector to small
And the detection performance of the target of dense distribution；Secondly, batch normalization is added in a network, data is made to become mean value 0, standard deviation
For 1 distribution；Also, promote learning rate；Deep neural network model is trained.
4. the method for deep learning air remote sensing target detection according to claim 1, which is characterized in that described in step 3
Test process includes:
Using the deep neural network constructed in the Airborne Data Classification collection and step 2 built in step 1, carry out end to end
Test:
1) picture to be detected is inputted, picture is input in sorter network to obtain various sizes of Feature Mapping；
2) wherein six layers of characteristic pattern is extracted, each point 6 different scale of construction respectively on these characteristic patterns are big later
Small encirclement frame, is then detected and is classified respectively, and multiple encirclement frames are generated；
3) different characteristic figure encirclement frame obtained is removed a part overlapping or not just with non-maxima suppression method
True encirclement frame generates final encirclement frame set；
4) result is evaluated and tested, FN: is judged as negative sample, but be in fact positive sample；FP: being judged as positive sample, but
It in fact is negative sample；TN: being judged as negative sample, in fact and negative sample；TP: it is judged as positive sample, in fact
It is card sample；It calculates accuracy rate P=TP/ (TP+FP), calculates recall rate R=TP/ (TP+FN).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910757510.6A CN110503020A (en) | 2019-08-16 | 2019-08-16 | A method of based on deep learning air remote sensing target detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910757510.6A CN110503020A (en) | 2019-08-16 | 2019-08-16 | A method of based on deep learning air remote sensing target detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110503020A true CN110503020A (en) | 2019-11-26 |
Family
ID=68588140
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910757510.6A Pending CN110503020A (en) | 2019-08-16 | 2019-08-16 | A method of based on deep learning air remote sensing target detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110503020A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111144335A (en) * | 2019-12-30 | 2020-05-12 | 自然资源部国土卫星遥感应用中心 | Method and device for building deep learning model |
| CN116363435A (en) * | 2023-04-03 | 2023-06-30 | 盐城工学院 | Remote sensing image target detection system and method based on deep learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108710897A (en) * | 2018-04-24 | 2018-10-26 | 江苏科海智能系统有限公司 | A kind of online general target detecting system in distal end based on SSD-T |
| CN109492674A (en) * | 2018-10-19 | 2019-03-19 | 北京京东尚科信息技术有限公司 | The generation method and device of SSD frame for target detection |
| CN110033417A (en) * | 2019-04-12 | 2019-07-19 | 江西财经大学 | A kind of image enchancing method based on deep learning |
-
2019
- 2019-08-16 CN CN201910757510.6A patent/CN110503020A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108710897A (en) * | 2018-04-24 | 2018-10-26 | 江苏科海智能系统有限公司 | A kind of online general target detecting system in distal end based on SSD-T |
| CN109492674A (en) * | 2018-10-19 | 2019-03-19 | 北京京东尚科信息技术有限公司 | The generation method and device of SSD frame for target detection |
| CN110033417A (en) * | 2019-04-12 | 2019-07-19 | 江西财经大学 | A kind of image enchancing method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| HONGYI ZHANG ET AL.: "mixup: BEYOND EMPIRICAL RISK MINIMIZATION", 《ICLR 2018》 * |
| 赵丹新 等: "基于ResNet的遥感图像飞机目标检测新方法", 《电子设计工程》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111144335A (en) * | 2019-12-30 | 2020-05-12 | 自然资源部国土卫星遥感应用中心 | Method and device for building deep learning model |
| CN116363435A (en) * | 2023-04-03 | 2023-06-30 | 盐城工学院 | Remote sensing image target detection system and method based on deep learning |
| CN116363435B (en) * | 2023-04-03 | 2023-10-27 | 盐城工学院 | Remote sensing image target detection system and method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108460382B (en) | Optical remote sensing image ship detection method based on deep learning single-step detector | |
| CN108038445B (en) | SAR automatic target identification method based on multi-view deep learning framework | |
| CN106127204B (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
| CN110889324A (en) | Thermal infrared image target identification method based on YOLO V3 terminal-oriented guidance | |
| CN112395987B (en) | SAR image target detection method based on unsupervised domain adaptive CNN | |
| CN110135522B (en) | Intelligent method for detecting and marking small target integration of remote sensing image | |
| CN108257154B (en) | Polarization SAR image change detection method based on regional information and CNN | |
| CN110189304A (en) | Remote sensing image target on-line quick detection method based on artificial intelligence | |
| CN105787501A (en) | Vegetation classification method capable of automatically selecting features in power transmission line corridor area | |
| CN112883971A (en) | SAR image ship target detection method based on deep learning | |
| CN106845343B (en) | Automatic detection method for optical remote sensing image offshore platform | |
| CN110503020A (en) | A method of based on deep learning air remote sensing target detection | |
| CN116563726A (en) | Remote sensing image ship target detection method based on convolutional neural network | |
| CN111192240B (en) | Remote sensing image target detection method based on random access memory | |
| CN112348758A (en) | Optical remote sensing image data enhancement method and target identification method | |
| CN116486243A (en) | DP-ViT-based sonar image target detection method | |
| Zhang et al. | Nearshore vessel detection based on Scene-mask R-CNN in remote sensing image | |
| Li et al. | Insect detection and counting based on YOLOv3 model | |
| CN102663773A (en) | Dual-core type adaptive fusion tracking method of video object | |
| CN117689995A (en) | Unknown spacecraft level detection method based on monocular image | |
| Zou et al. | Maritime target detection of intelligent ship based on faster R-CNN | |
| Zhang et al. | An improved YOLO algorithm for rotated object detection in remote sensing images | |
| CN110826575A (en) | Underwater target identification method based on machine learning | |
| Li et al. | SAR image object detection based on improved cross-entropy loss function with the attention of hard samples | |
| CN106997460A (en) | SAR target identification methods based on depth curve ripple convolutional network |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191126 |
|
| WD01 | Invention patent application deemed withdrawn after publication |