CN115546638A - Change detection method based on Siamese cascade differential neural network - Google Patents

Change detection method based on Siamese cascade differential neural network Download PDF

Info

Publication number
CN115546638A
CN115546638A CN202211219314.1A CN202211219314A CN115546638A CN 115546638 A CN115546638 A CN 115546638A CN 202211219314 A CN202211219314 A CN 202211219314A CN 115546638 A CN115546638 A CN 115546638A
Authority
CN
China
Prior art keywords
feature
change detection
graph
neural network
siamese
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
Application number
CN202211219314.1A
Other languages
Chinese (zh)
Inventor
江凯旋
刘嘉
张文华
刘芳
李东徽
王宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN202211219314.1A priority Critical patent/CN115546638A/en
Publication of CN115546638A publication Critical patent/CN115546638A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/176—Urban or other man-made structures
    • 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
    • G06V10/52—Scale-space analysis, e.g. wavelet analysis
    • 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
    • G06V10/77—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
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted 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
    • G06V10/82—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
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • 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)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a change detection method based on a Siamese differential neural network, which comprises the following steps: constructing a symmetrical bilateral cascade neural network structure; establishing a change detection learning model aiming at the reinforced feature extraction; optimizing the learning model based on an attention mechanism; and extracting an optimization result and outputting a change area. The invention has the advantages that a Siamese change detection model aiming at the change detection problem is defined, and the false detection caused by the alignment error of the double-time-state image can be well relieved, so that the invention has stronger robustness to the scale and color change in the image and can adapt to various data types.

Description

Change detection method based on Siamese cascade differential neural network
Technical Field
The invention belongs to the field of change detection, and particularly relates to a change detection method based on a Siamese cascade differential neural network.
Background
The change detection is to detect a changed region of the same scene image taken at different times. It is of great interest in many applications, including video monitoring, medical diagnosis and treatment, particularly in remote monitoring and land use analysis. Dual temporal CD is an integrated analysis of an image, where each point in the image is labeled as two groups: and the modified class and the unmodified class can obtain a final difference image and a CD result through analysis.
The conventional change detection method is to first perform image preprocessing such as geometric alignment and radiation correction; then generating pixel level differences by comparing features extracted from the double-temporal images; and then, carrying out image segmentation to divide the pixels into variable pixels and invariable pixels. However, these methods are highly dependent on expert experience and have many parameters determined manually, which makes it difficult to process large amounts of data with high accuracy. Thus, there are some limitations at the application level. With the rapid development of deep learning methods, many neural network models and components are used in the CD field to extract deeper feature representations, which makes it possible to extract feature maps by an end-to-end method. Although both CD methods have been practically successful, some problems remain. Due to the influence of irrelevant inconsistencies such as illumination change and alignment error, many existing methods have difficulty in accurately capturing spatial local information of an image. It can lead to uncertainty of pixel edges and false positives on objects.
Given two images, most co-registration methods take the captured scene as a plane and transform the images using a fixed transformation template, such as shift, rotation, and affine transformation. Therefore, for high resolution, images captured from different angles are difficult to align perfectly. Such images are common in many situations, such as Very High Resolution (VHR) optical remote sensing images and those captured by a moving drone (UVA). Therefore, in many change detection scenarios, a method that is robust against co-registration errors is needed. The target-based change detection method first classifies objects in the image and then compares the objects, which is robust to co-registration errors. In order to generate accurate change regions, a classification method should be specially designed. The intuitive approach is to classify the multiple time images separately, then compare the corresponding classes, and generate the changed regions. The accuracy of these methods depends on the accuracy of the classification method, respectively, and there is error propagation. Even with high self-matching and robustness to co-registration errors, object-based methods are typically supervised for learning accurate classifiers.
Disclosure of Invention
The invention discloses a change detection method based on a Siamese cascade differential neural network, which aims to detect changes in different application scenes, provides a change detection network based on a bilateral depth structure, and acquires change information through an attention mechanism and fusion characteristics.
The technical solution for realizing the purpose of the invention is as follows: in a first aspect, the present invention provides a change detection method based on a siamese differential neural network, including:
the first step, a symmetrical Siamese cascade neural network structure is constructed for learning, two registered pictures are input at two ends of the network, and characteristic graphs of different levels are obtained through a plurality of layers of convolution layers respectively;
secondly, defining a change detection learning model aiming at the reinforced feature extraction, and introducing an attention mechanism optimization feature graph;
thirdly, performing fusion upsampling on the feature by jumping connection on the feature graph subjected to the attention mechanism to generate feature graphs of different layers; further strengthening the characteristic graph by weighting, obtaining a corresponding difference graph by differentiating Euclidean distances, upsampling the difference graphs with different sizes to the same size, and finally adding the different difference graphs to obtain a final variation graph;
and fourthly, introducing a contrast loss function as a measure between the change graph and the real label graph, and training the neural network by minimizing the loss.
In a second aspect, the present application further provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the method according to the first aspect.
In a third aspect, the present application also provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method of the first aspect described above.
In a fourth aspect, the present application also provides a computer program product comprising a computer program which, when executed by a processor, implements the method of the first aspect described above.
Compared with the prior art, the invention has the remarkable characteristics that: (1) defining a bilateral convolutional neural network architecture; (2) Defining a multi-scale difference attention module and a weighted difference fusion mechanism module, and learning network parameters by using two input images; (3) Optimizing a loss function, balancing contrast loss by using batches, and aiming at modifying the weights of different loss items in the loss function so as to achieve a balance effect.
The invention utilizes two input images for learning network parameters in a supervised manner. The neural network consists of two parallel encoders with shared weights for extracting multi-scale features and a decoder for decoding the varying information by combining different differential feature maps. The ablation experiment proves the effectiveness of the multi-scale difference attention module and the weighted difference fusion mechanism module, which can obtain the corresponding relation between different characteristics and the intermediate semantic change graphs of different levels. The two modules have good effects in the aspects of image denoising, compression artifact suppression and the like. The method has the advantages that the method is proved by the fact that the method is verified under different data types and scenes, and the robustness of the co-registration error is improved.
The present invention is described in further detail below with reference to the attached drawing figures.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a diagram of a content encoding network according to the present invention.
FIG. 3 is a graph of the results of different methods on a data set.
Detailed Description
The following detailed description of the implementation of the present invention, with reference to fig. 1, includes the following steps:
firstly, a symmetrical bilateral network structure is constructed for extracting features, as shown in fig. 2, two registered pictures are input at two ends of the network, and feature maps of different levels are obtained through convolution of a plurality of layers respectively.
And a second step of defining a bilateral cascade neural network model. The specific process is as follows:
(1) Constructing a bilateral cascade change detection learning model, and defining input data as X and Y, and then defining a characteristic function as follows:
X i ,Y i ＝f i (X 1 ,Y 1 )
wherein X i ,Y i Two change detection images respectively representing inputs, f i (X 1 ,Y 1 ) Representing the obtained characteristic diagram, wherein i is the number of layers;
(2) And (3) introducing an attention mechanism optimization learning model, wherein the attention mechanism comprises the following steps:
(1) performing convolution on the input feature map by three different 1x1 to obtain three feature vectors for generating attention weight, wherein the three feature vectors are respectively expressed by Q, K and V;
(2) adjusting the dimensions of the three vectors of Q, K and V into a two-dimensional matrix, and respectively using the three vectors
Figure BDA0003876569440000031
To represent;
(3) will be provided with
Figure BDA0003876569440000032
Vector sum
Figure BDA0003876569440000033
Multiplying the transposed matrix, and obtaining a weight coefficient A through Softmax;
(4) normalization using score, i.e. divided by root number C';
Figure BDA0003876569440000034
wherein K is obtained by dimension conversion
Figure BDA0003876569440000035
Then obtaining the product through matrix transposition
Figure BDA0003876569440000036
Q is obtained by dimension conversion
Figure BDA0003876569440000037
Figure BDA0003876569440000038
And
Figure BDA0003876569440000039
the number of the characteristic channels is the same; c' represents the number of the characteristic channels;
(5) multiplying the matrix A by
Figure BDA0003876569440000041
Obtaining a weighted vector matrix for each input vector
Figure BDA0003876569440000042
Figure BDA0003876569440000043
(6) Will be provided with
Figure BDA0003876569440000044
The vector reshape is a three-dimensional feature vector and is represented by Y;
Y＝F(X)
(7) adding to obtain a final output result Z;
Z＝F(X)+X
wherein, Z is the output feature map, F (X) is the residual mapping function, which represents the similarity relation between each feature vector, and it is a trainable parameter, and the goal of the mechanism is to generate a weight coefficient distributed with the data in the whole data space.
And thirdly, fusing and upsampling the features of the feature map subjected to the attention mechanism through skip connection to generate feature maps of different layers. And then further strengthening the characteristic graph by weighting, then obtaining a corresponding difference graph by differentiating according to Euclidean distances, upsampling the difference graphs with different sizes to the same size, and finally adding the different difference graphs to obtain a final change graph.
Firstly, the feature map passing through the attention mechanism is up-sampled by jump connection to obtain four pairs of feature maps, and DX is used at the time of T1 i Indicating that DYi is used at time T2 i Showing that each feature map is subjected to channel transformation through convolution of 1 multiplied by 1 to generate a corresponding weight matrix W i W is to be i Multiplying the corresponding feature map to obtain weighted feature, and using DX for the weighted feature map i And DY i And (4) showing.
And then, performing Euclidean distance difference processing on the feature maps of different levels, wherein the formula is as follows:
DI i ＝E(DX i ,DY i )i＝1,2,3,4
wherein E (-) represents Euclidean distance, DI i And representing the difference graph obtained by Euclidean distance of different obtained hierarchies. Then, the feature maps of different levels are up-sampled to the same size, and the formula is as follows:
D i ＝upsample(DI i )i＝1,2,3,4
wherein D i And representing four groups of difference graphs after upsampling, and then adding the four groups of difference graphs to obtain a change graph of the final network output. The formula is as follows:
D＝D 1 +D 2 +D 3 +D 4
and fourthly, introducing a contrast loss function as a measure between the change graph and the real label graph, and training the neural network by minimizing the loss, wherein the loss function is defined as follows:
Figure BDA0003876569440000045
wherein D represents a prediction result graph, M represents a binarization label, wherein M represents changed when being 1 and unchanged when being 0, and subscripts b, i and j respectively represent batchsize, height and width; m is a boundary value limiting the size of the range of varying pixel pairs and is set to 2 in the experiment, where P u As the weight of the unchanged pixel pair, P c Representing the weights of the changed pixel pairs, which are computed from the label values of the corresponding classes.
P u ＝pos_num
P c ＝neg_num
Pos _ num and neg _ num respectively represent the number of unchanged pixel pairs and changed pixel pairs, and are calculated according to the label values of the corresponding categories.
Unlike previous separately encoded bi-temporal images, the present invention designs an encoder difference attention module to focus on the spatial difference relationship of pixels. To improve the generalization ability of the network, it calculates attention weights between arbitrary pixels between the bi-temporal images and uses them to produce more discriminative features. In order to improve feature fusion and avoid gradient disappearance, a multi-scale weighted variance mapping fusion strategy is provided in a decoding stage to obtain a better change detection result, so that the method has stronger adaptability.
The effect of the invention can be further illustrated by the following simulation experiment:
simulation conditions
Simulation experiment seasonal variation detection data set: the data set comprises 16000 changing season Google earth images with pixel-by-pixel change detection labels, 3000 test samples, 3000 verification samples and 10000 training samples, and can provide information of changes of common objects such as buildings and land and change information of a plurality of detailed objects such as automobiles and roads. Here, we use this dataset to verify the feasibility of the proposed probabilistic model in change detection and compare the capabilities of four unsupervised change detection methods in processing image details, simulation experiments were performed in both AMD Ryzen 5600X (3.7 GHz) and RTX 3060 GPUs in a configuration environment under Windows 11 operating system, and programs were written using Python and PyCharm 2021.
The evaluation indexes adopted by the invention are accuracy (Precision), recall (Recall), F1 Score (F1-Score), average cross-over ratio (IOU, interaction-over-Union)
Emulated content
The invention uses seasonal variations to detect the performance of the data set inspection algorithm. In order to test the performance of the algorithm, the change detection method based on the Siamese cascade differential neural network is compared with the change detection algorithm popular in the world at present. The comparison method comprises the following steps: FC-EF, FC-Sim-diff, FC-Sim-conc, STANet.
Analysis of simulation experiment results
Table 1 shows the comparison results of different evaluation indexes under different change detection algorithms for two data sets, and it can be seen from table 1 that, in the S1 data set, the change detection method based on the siamenon cascade differential neural network proposed by the present invention has strong adaptability, can highlight the changed area, avoid the influence of unchanged building, and significantly improve the precision in different evaluation indexes compared with FC-EF, FC-Siam-diff, FC-Siam-conc, STANet. The graph of the effect of the results of the inventive method and the comparative method on the change detection data set is shown in fig. 3. The simulation experiment result of the real data set shows the effectiveness of the method.
TABLE 1 quantitative evaluation of different algorithms of the data set
Figure BDA0003876569440000061
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, and these are all within the scope of protection of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims (8)

1. A change detection method based on a Siamese differential neural network is characterized by comprising the following steps:
the method comprises the following steps that firstly, a symmetrical Siamese cascade neural network structure is constructed for learning, two registered pictures are input at two ends of the network, and feature maps of different levels are obtained through a plurality of layers of convolution layers respectively;
secondly, defining a change detection learning model aiming at the reinforced feature extraction, and introducing an attention mechanism optimization feature map;
thirdly, performing fusion upsampling on the feature by jumping connection on the feature graph subjected to the attention mechanism to generate feature graphs of different layers; further strengthening the characteristic graph by weighting, obtaining a corresponding difference graph by differentiating Euclidean distances, upsampling the difference graphs with different sizes to the same size, and finally adding the different difference graphs to obtain a final variation graph;
and fourthly, introducing a contrast loss function as a measure between the change graph and the real label graph, and training the neural network by minimizing the loss.
2. The Siamese differential neural network-based change detection method of claim 1, wherein the second step comprises the following specific steps:
(1) Constructing a bilateral cascade change detection learning model, and defining input data as X and Y, so that defining a characteristic function as follows:
X i ,Y i ＝f i (X 1 ,Y 1 )
wherein X i ,Y i Two change detection images respectively representing inputs, f i (X 1 ,Y 1 ) Representing the obtained characteristic diagram, wherein i is the number of layers;
(2) And (3) introducing an attention mechanism optimization learning model, wherein the attention mechanism comprises the following steps:
(1) performing convolution on the input feature diagram by three different 1x1 to obtain three feature vectors, wherein the three feature vectors are used for generating attention weight and are respectively expressed by Q, K and V;
(2) adjusting the dimensions of the three vectors of Q, K and V into a two-dimensional matrix, and respectively using the three vectors
Figure FDA0003876569430000011
To represent;
(3) will be provided with
Figure FDA0003876569430000012
Vector sum
Figure FDA0003876569430000013
Multiplying the transposed matrix, and obtaining a weight coefficient A through Softmax;
(4) normalization using score;
Figure FDA0003876569430000014
wherein K is obtained by dimension conversion
Figure FDA0003876569430000015
Then obtaining the product through matrix transposition
Figure FDA0003876569430000016
Q is obtained by dimension conversion
Figure FDA0003876569430000017
Figure FDA0003876569430000018
And
Figure FDA0003876569430000019
the number of the characteristic channels is the same; c' represents the number of the characteristic channels;
(5) will momentMultiplication of array A by
Figure FDA0003876569430000021
Obtaining a weighted vector matrix for each input vector
Figure FDA0003876569430000022
Figure FDA0003876569430000023
(6) Will be provided with
Figure FDA0003876569430000024
The vector reshape is a three-dimensional feature vector and is represented by Y;
Y＝F(X)
(7) adding to obtain a final output result Z;
Z＝F(X)+X
where Z is the output feature map and F (X) is the residual mapping function, representing the similarity relationship between each feature vector.
3. The Siamese differential neural network-based change detection method of claim 2, wherein in the third step, the feature map subjected to the attention mechanism is subjected to fusion upsampling on the features through jump connection to generate feature maps of different levels; then, further strengthening the characteristic graph by weighting, then obtaining a corresponding difference graph by performing difference on Euclidean distances, upsampling the difference graphs with different sizes to the same size, and finally adding the different difference graphs to obtain a final change graph;
firstly, the feature map passing through the attention mechanism is up-sampled by jump connection to obtain four pairs of feature maps, and DX is used at the time of T1 i Indicating that DY is used at time T2 i Showing that each feature map is subjected to channel transformation through convolution of 1 multiplied by 1 to generate a corresponding weight matrix W i W is to be i Multiplying the corresponding feature map to obtain weighted features, and using D to obtain the weighted feature mapX i And DY i Represents;
and then, performing Euclidean distance difference processing on the feature maps of different levels, wherein the formula is as follows:
DI i ＝E(DX i ,DY i )i＝1,2,3,4
wherein E (-) denotes Euclidean distance, DI i Representing a difference graph obtained by Euclidean distance of different obtained layers; then, the feature maps of different levels are up-sampled to the same size, and the formula is as follows:
D i ＝upsample(DI i )i＝1,2,3,4
wherein D i Representing four groups of difference graphs after upsampling, and then adding the four groups of difference graphs to obtain a change graph of the final network output, wherein the formula is as follows:
D＝D 1 +D 2 +D 3 +D 4 。
4. the Siamese differential neural network-based change detection method of claim 1, wherein the fourth step introduces a comparison loss function as a measure between the change map and the real label map, trains the neural network by minimizing loss, and defines a loss function as follows:
Figure FDA0003876569430000025
wherein D represents a prediction result graph, M represents a binarization label, wherein M represents changed when being 1 and unchanged when being 0, and subscripts b, i and j respectively represent batchsize, height and width; m is a boundary value for limiting the range size of the changed pixel pair; p u As the weight of the unchanged pixel pair, P c Representing the weights of the changed pixel pairs, which are computed from the label values of the respective classes.
P u ＝pos_num
P c ＝neg_num
Pos _ num and neg _ num respectively represent the number of unchanged pixel pairs and changed pixel pairs, and are calculated according to the label values of the corresponding categories.
5. The Siamese differential neural network-based change detection method of claim 4, wherein m =2.
6. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the method of any of claims 1-5 are implemented when the program is executed by the processor.
7. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 5.
8. A computer program product comprising a computer program, characterized in that the computer program realizes the steps of the method of any one of claims 1-5 when executed by a processor.
CN202211219314.1A 2022-09-30 2022-09-30 Change detection method based on Siamese cascade differential neural network Pending CN115546638A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211219314.1A CN115546638A (en) 2022-09-30 2022-09-30 Change detection method based on Siamese cascade differential neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211219314.1A CN115546638A (en) 2022-09-30 2022-09-30 Change detection method based on Siamese cascade differential neural network

Publications (1)

Publication Number Publication Date
CN115546638A true CN115546638A (en) 2022-12-30

Family

ID=84731025

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211219314.1A Pending CN115546638A (en) 2022-09-30 2022-09-30 Change detection method based on Siamese cascade differential neural network

Country Status (1)

Country Link
CN (1) CN115546638A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117132554A (en) * 2023-08-10 2023-11-28 佳源科技股份有限公司 Transformer substation equipment defect detection method, system, computer equipment and storage medium
CN117726614A (en) * 2023-12-28 2024-03-19 徐州医科大学 Quality perception network and attention-like Siamese network collaborative medical fusion image quality evaluation method

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117132554A (en) * 2023-08-10 2023-11-28 佳源科技股份有限公司 Transformer substation equipment defect detection method, system, computer equipment and storage medium
CN117726614A (en) * 2023-12-28 2024-03-19 徐州医科大学 Quality perception network and attention-like Siamese network collaborative medical fusion image quality evaluation method

Similar Documents

Publication Publication Date Title
CN110135267B (en) Large-scene SAR image fine target detection method
Nandhini Abirami et al. Deep CNN and Deep GAN in Computational Visual Perception‐Driven Image Analysis
CN105678284B (en) A kind of fixed bit human body behavior analysis method
CN110059728B (en) RGB-D image visual saliency detection method based on attention model
CN115546638A (en) Change detection method based on Siamese cascade differential neural network
CN109191424B (en) Breast mass detection and classification system and computer-readable storage medium
CN111325750B (en) Medical image segmentation method based on multi-scale fusion U-shaped chain neural network
CN116563285B (en) Focus characteristic identifying and dividing method and system based on full neural network
CN114677722A (en) Multi-supervision human face in-vivo detection method integrating multi-scale features
CN116524189A (en) High-resolution remote sensing image semantic segmentation method based on coding and decoding indexing edge characterization
CN115861756A (en) Earth background small target identification method based on cascade combination network
CN114048810A (en) Hyperspectral image classification method based on multilevel feature extraction network
Hughes et al. A semi-supervised approach to SAR-optical image matching
CN117576724A (en) Unmanned plane bird detection method, system, equipment and medium
Zuo et al. A remote sensing image semantic segmentation method by combining deformable convolution with conditional random fields
CN116977747B (en) Small sample hyperspectral classification method based on multipath multi-scale feature twin network
CN116129280B (en) Method for detecting snow in remote sensing image
Bakır et al. Evaluating the robustness of yolo object detection algorithm in terms of detecting objects in noisy environment
CN117576461A (en) Semantic understanding method, medium and system for transformer substation scene
CN117315473A (en) Strawberry maturity detection method and system based on improved YOLOv8
CN107358625A (en) The High Resolution SAR image change detection method detected based on SPP Net and area-of-interest
CN116953702A (en) Rotary target detection method and device based on deduction paradigm
CN114821074B (en) Airborne LiDAR point cloud semantic segmentation method, electronic equipment and storage medium
CN108154107B (en) Method for determining scene category to which remote sensing image belongs
Kurdthongmee et al. A framework to create a deep learning detector from a small dataset: A case of parawood pith estimation

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