CN116958809A - Remote sensing small sample target detection method for feature library migration - Google Patents
Remote sensing small sample target detection method for feature library migration Download PDFInfo
- Publication number
- CN116958809A CN116958809A CN202310717317.6A CN202310717317A CN116958809A CN 116958809 A CN116958809 A CN 116958809A CN 202310717317 A CN202310717317 A CN 202310717317A CN 116958809 A CN116958809 A CN 116958809A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- target
- small sample
- detection model
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 78
- 238000013508 migration Methods 0.000 title claims abstract description 11
- 230000005012 migration Effects 0.000 title claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 21
- 238000000034 method Methods 0.000 claims abstract description 14
- 239000011159 matrix material Substances 0.000 claims abstract description 11
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 10
- 238000000605 extraction Methods 0.000 claims abstract description 10
- 239000013598 vector Substances 0.000 claims description 38
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims 1
- 230000006870 function Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
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
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/766—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using regression, e.g. by projecting features on hyperplanes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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
Abstract
The application discloses a method for detecting a remote sensing small sample target by feature library migration, which comprises the steps of firstly, aiming at an input remote sensing image to be detected, carrying out multi-scale feature extraction based on a deep convolutional neural network to generate a detection model of a common category; setting a small sample feature library, adding the small sample feature library into a detection model for training, and enabling the detection model to maintain the detection capability of common targets while detecting small samples; aiming at the characteristic of high symmetry of the remote sensing image to be detected, when the detection result is predicted after the detection model is trained, the parameter coefficient of the symmetric matrix is introduced, and the detection precision of the small sample detection target is improved by correcting the positioning branch parameter. The method solves the problems that modeling is poor and accurate identification cannot be performed in the remote sensing image aiming at the target class characteristics with a relatively rare number, and improves the effectiveness of remote sensing image detection in practical application.
Description
Technical Field
The application relates to the technical field of target detection, in particular to a remote sensing small sample target detection method for feature library migration.
Background
The object detection problem is an important problem in the field of remote sensing image processing, wherein the detection of a small sample object is a difficult task scene, but has extremely high application value. The small sample target detection is a task of accurately positioning and identifying unusual or scarce target types in a scene, and the small sample target detection technology is used for extracting characteristic information of targets from a data set consisting of common type samples and a small number of scarce type samples by learning to construct a model capable of accurately identifying and positioning the common type targets and accurately identifying and positioning the scarce type targets.
In recent years, the small sample detection technology has been developed rapidly, and there are many excellent small sample target detection algorithms, and for targets with relatively large target dimensions in images (the aspect ratio of the images and the targets is generally not more than 10) and relatively small numbers, the algorithms can realize relatively good performance under the condition of small samples, but for remote sensing images, due to the characteristics of large image size, large aspect ratio of the images and the targets, wide target size change range and the like, the conventional target detection technology has poor effect when applied to such small sample detection tasks with only a small number of samples, and a detection model cannot accurately detect and identify the targets through a small number of training samples.
Disclosure of Invention
The application aims to provide a remote sensing small sample target detection method for feature library migration, which solves the problems that modeling is poor and accurate identification cannot be realized aiming at a few target class features in a remote sensing image, and improves the effectiveness of remote sensing image detection in practical application.
The application aims at realizing the following technical scheme:
a method for detecting a target of a remote sensing small sample for feature library migration, the method comprising:
step 1, firstly, aiming at an input remote sensing image to be detected, carrying out multi-scale feature extraction based on a deep convolutional neural network, and generating a detection model of a non-small sample type;
step 2, setting a small sample feature library, and adding the small sample feature library into a detection model for training, so that the detection model can maintain the detection capability of common targets while detecting small samples;
and 3, introducing a symmetric matrix parameter coefficient when the detection model is trained and predicting the detection result according to the characteristic of high symmetry of the remote sensing image to be detected, and correcting the regression branch parameter output by the prediction result of the detection model to improve the detection precision of the small sample detection target.
According to the technical scheme provided by the application, the problems that modeling is poor and accurate identification cannot be performed in the remote sensing image aiming at the target class characteristics with a relatively rare number are solved, and the effectiveness of remote sensing image detection in practical application is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of a remote sensing small sample target detection method for feature library migration provided by an embodiment of the application.
Detailed Description
The technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments of the present application, and this is not limiting to the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to fall within the scope of the application.
Fig. 1 is a schematic flow chart of a remote sensing small sample target detection method for feature library migration according to an embodiment of the present application, where the method includes:
step 1, firstly, aiming at an input remote sensing image to be detected, carrying out multi-scale feature extraction based on a deep convolutional neural network, and generating a detection model of a non-small sample type;
in the step, a remote sensing data set A containing non-small sample type targets is firstly set, and position coordinates and type labels are carried out on the non-small sample type targets, so that no small sample targets or small sample targets in the remote sensing data set A do not participate in loss function calculation in training;
and setting a convolutional neural network based on FaterRCNN, setting a characteristic extraction module of the convolutional neural network as a high-resolution characteristic extraction module of HRNet, and training the convolutional neural network by using the remote sensing data set A.
Step 2, setting a small sample feature library, and adding the small sample feature library into a detection model for training, so that the detection model can maintain the detection capability of common targets while detecting small samples;
in the step, a support (support) branch is added into a trained detection model, a remote sensing data set B containing a small sample target is set, and position coordinates and category labels are carried out on the small sample target, so that no common target or common target in the remote sensing data set B is ensured to be not used as a negative sample to participate in loss function calculation in training;
fixing other parameters of the detection model, and independently training the support branches and the classifier to obtain the parameters of the feature extractor of the small sample target;
for each type of sample of the remote sensing data set B, a sample quantity K is set, and each sample is characterized by supporting branch generation
The vector is m, the characteristic vector generated by the sample through the detection model is n, and the average vectors of K samples are respectivelyConstructing a class feature library, carrying out similarity measurement on the class feature library and the class feature library, selecting cosine similarity, and taking the similarity beta generated by each small sample target as a parameter coefficient of a detection model branch;
the output of the prediction result of the detection model is divided into a classification branch for judging the target category and a regression branch for positioning the target position; removing 1*P-dimensional vectors of the last layer of the classification branch, wherein P is the number of common targets, Q is the number of small sample targets, initializing a layer of 1 (P+Q) vectors as classification vectors, performing model training on the remote sensing data sets A and B, and when the categories in the remote sensing data set A are trained, setting the parameter coefficient of the classification branch to be 1; when training the category in the remote sensing data set B, the coefficient of the classification branch is the corresponding similarity beta generated in 4;
after the detection model converges, setting the parameter coefficient of the classification branch to be 1, training again until the detection model converges, and finally finishing the detection model training.
And 3, introducing a symmetric matrix parameter coefficient when the detection model is trained and predicting the detection result according to the characteristic of high symmetry of the remote sensing image to be detected, and correcting the regression branch parameter output by the prediction result of the detection model to improve the detection precision of the small sample detection target.
In this step, the feature vectors of the remote sensing image to be detected are multiplied by a plurality of symmetric matrices, and then the similarity between each result and the original vector, namely the distance measurement, is compared, half of the included angle between the result and the minimum distance vector is taken as the target orientation, and the coefficient is added to the last layer of parameters of the detection model by adding a rotation matrix, specifically:
after removing the support branches, carrying out feature extraction and feature library vector comparison on the input remote sensing image to be detected to obtain vectors of a regression branch and a classification branch;
and decoding the vector of the regression branch, multiplying the vector with the following 4 matrixes according to the symmetry of the vector in the two-dimensional plane after decoding the vector back to the coordinate plane, performing cosine similarity calculation on the obtained 4 feature vectors and the original vector, taking half of the included angle between the feature vectors and the minimum distance vector as a target orientation, and marking the included angle as theta, wherein the 4 matrixes are as follows:
adding a rotation matrix according to the included angle theta to add coefficients for the last layer of parameters of the detection model, wherein the rotation matrix is as follows:
finally, carrying out softmax classification through the prediction result of the classified branch to obtain a classification result; and performing positioning prediction through regression branches.
It is noted that what is not described in detail in the embodiments of the present application belongs to the prior art known to those skilled in the art.
In conclusion, the method disclosed by the embodiment of the application can better detect the small sample type and the common type in the small sample target detection task, has better performance than the existing various algorithms, and has far better performance than other methods in the remote sensing image; aiming at the characteristic of high symmetry of the remote sensing image target, the application provides accurate positioning and classification supervision by introducing a symmetrical matrix transformation mode, and further improves the accuracy of remote sensing small target detection.
In addition, the method is compared with the current mainstream small sample target detection algorithm, the accuracy in the remote sensing image scene is greatly improved, and the model detection performance is superior to that of the existing algorithm under the condition that the sample size of each small sample target is single sample, three samples and five samples, so that the effectiveness of the method is proved.
In addition, it will be understood by those skilled in the art that all or part of the steps in implementing the methods of the above embodiments may be implemented by a program to instruct related hardware, and the corresponding program may be stored in a computer readable storage medium, where the storage medium may be a read only memory, a magnetic disk or an optical disk, etc.
The foregoing is only a preferred embodiment of the present application, but the scope of the present application is not limited thereto, and any changes or substitutions easily contemplated by those skilled in the art within the scope of the present application should be included in the scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims. The information disclosed in the background section herein is only for enhancement of understanding of the general background of the application and is not to be taken as an admission or any form of suggestion that this information forms the prior art already known to those of ordinary skill in the art.
Claims (4)
1. The method for detecting the target of the remote sensing small sample by migrating the feature library is characterized by comprising the following steps of:
step 1, firstly, aiming at an input remote sensing image to be detected, carrying out multi-scale feature extraction based on a deep convolutional neural network, and generating a detection model of a non-small sample type;
step 2, setting a small sample feature library, and adding the small sample feature library into a detection model for training, so that the detection model can maintain the detection capability of common targets while detecting small samples;
and 3, introducing a symmetric matrix parameter coefficient when the detection model is trained and predicting the detection result according to the characteristic of high symmetry of the remote sensing image to be detected, and correcting the regression branch parameter output by the prediction result of the detection model to improve the detection precision of the small sample detection target.
2. The method for detecting the remote sensing small sample target with the feature library migration according to claim 1, wherein in the step 1, firstly, a remote sensing data set A containing a non-small sample type target is set, and position coordinates and type labeling are carried out on the non-small sample type target, so that no small sample target or small sample target in the remote sensing data set A does not participate in loss function calculation in training;
and setting a convolutional neural network based on FaterRCNN, setting a characteristic extraction module of the convolutional neural network as a high-resolution characteristic extraction module of HRNet, and training the convolutional neural network by using the remote sensing data set A.
3. The method for detecting the small remote sensing sample target by the feature library migration according to claim 1, wherein in the step 2, a support branch is added into a trained detection model, a remote sensing data set B containing the small sample target is set, and the small sample target is marked with position coordinates and categories, so that no common target or common target in the remote sensing data set B is ensured not to be used as a negative sample to participate in loss function calculation in training;
fixing other parameters of the detection model, and independently training the support branches and the classifier to obtain the parameters of the feature extractor of the small sample target;
for each type of sample of the remote sensing data set B, the sample quantity is K, the feature vector generated by each sample through the support branch is m, the feature vector generated by the sample through the detection model is n, and the average vectors of the K samples are respectivelyConstructing a class feature library, carrying out similarity measurement on the class feature library and the class feature library, selecting cosine similarity, and taking the similarity beta generated by each small sample target as a parameter coefficient of a detection model branch;
the output of the prediction result of the detection model is divided into a classification branch for judging the target category and a regression branch for positioning the target position; removing 1*P-dimensional vectors of the last layer of the classification branch, wherein P is the number of common targets, Q is the number of small sample targets, initializing a layer of 1 (P+Q) vectors as classification vectors, performing model training on the remote sensing data sets A and B, and when the categories in the remote sensing data set A are trained, setting the parameter coefficient of the classification branch to be 1; when training the category in the remote sensing data set B, the coefficient of the classification branch is the corresponding similarity beta generated in 4;
after the detection model converges, setting the parameter coefficient of the classification branch to be 1, training again until the detection model converges, and finally finishing the detection model training.
4. The method for detecting a target of a small remote sensing sample with feature library migration according to claim 1, wherein in step 3, a plurality of symmetric matrices are multiplied on feature vectors of a remote sensing image to be detected, and then similarity between each result and an original vector, namely, a distance measure is compared, a half of an included angle with a minimum distance vector is taken as a target orientation, and a coefficient is added to a last layer of parameters of a detection model by adding a rotation matrix, specifically:
after removing the support branches, carrying out feature extraction and feature library vector comparison on the input remote sensing image to be detected to obtain vectors of a regression branch and a classification branch;
and decoding the vector of the regression branch, multiplying the vector with the following 4 matrixes according to the symmetry of the vector in the two-dimensional plane after decoding the vector back to the coordinate plane, performing cosine similarity calculation on the obtained 4 feature vectors and the original vector, taking half of the included angle between the feature vectors and the minimum distance vector as a target orientation, and marking the included angle as theta, wherein the 4 matrixes are as follows:
adding a rotation matrix according to the included angle theta to add coefficients for the last layer of parameters of the detection model, wherein the rotation matrix is as follows:
finally, carrying out softmax classification through the prediction result of the classified branch to obtain a classification result; and performing positioning prediction through regression branches.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310717317.6A CN116958809A (en) | 2023-06-16 | 2023-06-16 | Remote sensing small sample target detection method for feature library migration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310717317.6A CN116958809A (en) | 2023-06-16 | 2023-06-16 | Remote sensing small sample target detection method for feature library migration |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116958809A true CN116958809A (en) | 2023-10-27 |
Family
ID=88457374
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310717317.6A Pending CN116958809A (en) | 2023-06-16 | 2023-06-16 | Remote sensing small sample target detection method for feature library migration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116958809A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117409340A (en) * | 2023-12-14 | 2024-01-16 | 上海海事大学 | Unmanned aerial vehicle cluster multi-view fusion aerial photography port monitoring method, system and medium |
-
2023
- 2023-06-16 CN CN202310717317.6A patent/CN116958809A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117409340A (en) * | 2023-12-14 | 2024-01-16 | 上海海事大学 | Unmanned aerial vehicle cluster multi-view fusion aerial photography port monitoring method, system and medium |
CN117409340B (en) * | 2023-12-14 | 2024-03-22 | 上海海事大学 | Unmanned aerial vehicle cluster multi-view fusion aerial photography port monitoring method, system and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Xie et al. | Multilevel cloud detection in remote sensing images based on deep learning | |
CN110991311B (en) | Target detection method based on dense connection deep network | |
CN111079780B (en) | Training method for space diagram convolution network, electronic equipment and storage medium | |
CN113240012B (en) | Two-dimensional image-based unsupervised multi-view three-dimensional target retrieval method and device | |
Yang et al. | Active matting | |
CN116503399B (en) | Insulator pollution flashover detection method based on YOLO-AFPS | |
CN110942473A (en) | Moving target tracking detection method based on characteristic point gridding matching | |
CN111738319B (en) | Clustering result evaluation method and device based on large-scale samples | |
CN116958809A (en) | Remote sensing small sample target detection method for feature library migration | |
Kang et al. | Noise-tolerant deep neighborhood embedding for remotely sensed images with label noise | |
CN112364881A (en) | Advanced sampling consistency image matching algorithm | |
CN111339342B (en) | Three-dimensional model retrieval method based on angle ternary center loss | |
CN111144466B (en) | Image sample self-adaptive depth measurement learning method | |
CN114155388B (en) | Image recognition method and device, computer equipment and storage medium | |
CN113011597B (en) | Deep learning method and device for regression task | |
CN111695526B (en) | Network model generation method, pedestrian re-recognition method and device | |
Wang et al. | VTAAN: Visual tracking with attentive adversarial network | |
CN114241470A (en) | Natural scene character detection method based on attention mechanism | |
CN114463543A (en) | Weak supervision semantic segmentation method based on cascade decision and interactive annotation self-promotion | |
CN110263196B (en) | Image retrieval method, image retrieval device, electronic equipment and storage medium | |
CN113449631A (en) | Image classification method and system | |
CN112766465A (en) | Training method of neural network for intelligent rotation performance detection | |
JP2014092817A (en) | Character recognition device and program | |
CN116129206B (en) | Processing method and device for image decoupling characterization learning and electronic equipment | |
Perez-Montes et al. | An Efficient Facial Verification System for Surveillance that Automatically Selects a Lightweight CNN Method and Utilizes Super-Resolution Images |
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 |