CN117054754A - Quick radio storm signal searching method based on target detection model - Google Patents
Quick radio storm signal searching method based on target detection model Download PDFInfo
- Publication number
- CN117054754A CN117054754A CN202311024924.0A CN202311024924A CN117054754A CN 117054754 A CN117054754 A CN 117054754A CN 202311024924 A CN202311024924 A CN 202311024924A CN 117054754 A CN117054754 A CN 117054754A
- Authority
- CN
- China
- Prior art keywords
- model
- signal
- data
- frb
- rapid
- 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 23
- 238000012549 training Methods 0.000 claims abstract description 18
- 238000012360 testing method Methods 0.000 claims abstract description 9
- 238000013527 convolutional neural network Methods 0.000 claims description 12
- 230000000694 effects Effects 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000004088 simulation Methods 0.000 claims description 3
- 238000010200 validation analysis Methods 0.000 claims description 2
- 238000012795 verification Methods 0.000 claims 1
- 238000005336 cracking Methods 0.000 abstract description 13
- 239000002699 waste material Substances 0.000 abstract description 6
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 239000006185 dispersion Substances 0.000 description 14
- 238000004422 calculation algorithm Methods 0.000 description 6
- 230000008030 elimination Effects 0.000 description 5
- 238000003379 elimination reaction Methods 0.000 description 5
- 239000000463 material Substances 0.000 description 4
- 238000010586 diagram Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 241001620634 Roger Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R29/00—Arrangements for measuring or indicating electric quantities not covered by groups G01R19/00 - G01R27/00
- G01R29/02—Measuring characteristics of individual pulses, e.g. deviation from pulse flatness, rise time or duration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a rapid radio storm signal searching method based on a target detection model, and belongs to the technical field of astronomy and artificial intelligence. Aiming at the problems of large workload, simplicity and repetition, time waste caused by DM value violent cracking and the like when a rapid electric storm signal searching task is manually completed, a trainable target detection model is introduced to replace manual completion of rapid electric storm signal searching work, a training set and a testing set are divided into 1:1, model training is completed under the condition of less existing data, a better testing result is obtained, and DM value cracking speed and rapid electric storm signal searching efficiency are remarkably improved. The method has wide application prospect in astronomical signal searching and other fields.
Description
Technical Field
The invention relates to a rapid radio storm signal searching method based on a target detection model, and belongs to the technical field of astronomy and artificial intelligence.
Background
The fast radio storm (Fast Radio Bursts, FRB) is an explosive, pulsed radio radiation astronomical phenomenon of only a few milliseconds in duration, which is one of the current astronomical hotspot fronts. Because of the characteristics of extremely short burst time and few repetition, the rapid radio storm signal is difficult to capture.
Most of the existing fast radio storm signal searching methods firstly perform dispersion elimination treatment on signals, and the traditional technology generally adopts an automatic and high-performance software pipeline based on the dispersion elimination theory to perform FRB event searching, such as HEIMDALL, FDMT (Fast Dispersion Measure Transform), presto (Pulsar Exploration and Search Toolkit), BEAR (Burst Emission Automatic Roger) and the like. These algorithms face noise and RFI-induced false positive challenges, subject to radio frequency interference. In addition, the traditional searching method needs to perform dispersion elimination processing firstly, namely, the DM value needs to be known firstly, but the DM value is mostly cracked by adopting an exhaustion method at present, namely, experiments are performed by a traversing method, and the method is low in efficiency and wastes computational resources. With the rapid increase of the FRB observed data quantity, the number of false positive candidates is correspondingly increased, the workload of DM value calculation is very huge, and the screening work of FRB events from a large number of candidates is manually carried out, so that the search of rapid electric storm consumes manpower and material resources greatly.
Along with the development of artificial intelligence, a machine learning target detection technology is applied to the field of rapid electric storm, and the cycle-frequency diagram of the rapid electric storm without dispersion is directly learned, so that the current situation that a great deal of manpower is consumed in rapid electric storm searching is changed. The constructed convolutional neural network is used for carrying out regression to predict the DM value to replace violent cracking, so that the current situation that a large amount of material resources are consumed in the process of decoloring can be changed.
Currently, the transient non-relevant publications relate to the acquisition of a fast-shot storm signal by directly performing target detection on a fast-shot storm period-frequency map without dispersion. By adopting the traditional method, the quick electric storm searching is carried out on the basis of achromatizing, and for the DM regression problem, the violent cracking is adopted at present, and related publications such as:
fast radio storm candidate classification based on convolutional neural networks. And carrying out dispersion elimination processing and DM conversion on the dynamic spectrum data to respectively generate a dispersion elimination dynamic spectrum array and a DM-time intensity array. Finally input it into trainingThe convolutional neural network classifier is used for identifying and classifying, so that the classification of the rapid radio storm is completed, and the conventional violent cracking is adopted for the DM regression problem; the algorithm of the write channel brute force DM value as in Applying Deep Learning to Fast Radio Burst Classification requires that all N be t N of NDM DM test of time samples f And summing the frequency channels. The computational complexity is O (N) f N t log 2 N f ). The de-dispersion of trees employs a divide-by-conquer technique that exploits redundancy of adjacent channel de-dispersion (Taylor 1974). By using an FFT-like approach, the problem is reduced to having log 2 N f Branched tree, allowing O (N t N t log 2 N f ) Is not limited by the complexity of (a). A highly optimized CPU-based version of this algorithm has been used for the FRB search of the prime. Other algorithms, such as fast DM transforms (FDMT), exploit the same redundancy and attempt to maintain optimality. The algorithm is used for the FREDDA pipe of ASKAP.
The convolutional neural network is used for regression, the most classical CNN structure using the convolutional neural network is generally a plurality of convolutional layers, the convolutional layers are connected with full connection layers, and finally, the Softmax layer is connected with the output predicted classification probability. If the matrix of the image is also regarded as a vector, whether convolution or FC is performed in the CNN, one tensor is continuously transformed into another tensor, and finally the output is a probability vector with the number of categories formulated for classification as the dimension, and the value of the vector finally output is directly taken as regression, and the final optimized objective function is not cross entry and the like any more, but is directly based on the error of real values.
Disclosure of Invention
The invention aims at solving the problems and the defects existing in the prior art, and creatively provides a rapid shot storm signal searching method based on a target detection model, aiming at solving the technical problems of consumption of manpower and material resources, time waste caused by forced cracking of a DM value in the process of achromatizing in the process of firstly eliminating chromatic dispersion and then searching in the process of manually searching the rapid shot storm signal.
The method can effectively solve the problems of consumption of manpower and material resources, time waste caused by forced cracking of the DM value in the process of achromatizing in the process of manually searching the rapid electric storm signal, and the like, and can be used for applying target detection to the searching work of the rapid electric storm signal, thereby remarkably improving the cracking speed of the DM value and the searching efficiency of the rapid electric storm signal.
The innovation points of the invention include: aiming at the problems of large workload, simple repetition, time waste caused by DM value violent cracking and the like when a rapid electric storm signal searching task is manually completed, a trainable target detection model is introduced to replace manual completion of the rapid electric storm signal searching work, a training set and a testing set are divided into 1:1, model training is completed under the condition of less existing data, a good testing result is obtained, and the efficiency of the rapid electric storm signal searching work is improved. Aiming at the problem of time waste caused by the forced cracking of the DM value in the existing achromatizing process, a DM value regression model is introduced to replace manual completion of the searching work of the rapid radio storm signal, so that achromatizing efficiency is improved.
Advantageous effects
Compared with the prior art, the invention has the following characteristics:
firstly, the rapid radio storm detection is carried out on the non-dispersion signals which are directly captured, so that extremely time-consuming processes such as cracking of DM values and the like are avoided, a large amount of time and resources are saved, and the target detection technology is directly utilized to detect the non-dispersion time-frequency two-dimensional intensity map to judge whether the signals are rapid radio storm signals.
And secondly, carrying out DM value regression by using the constructed convolutional neural network to replace the traditional brute force cracking method of traversing the DM value in a certain step length within the range of 0-2000.
Drawings
FIG. 1 is a schematic flow chart of the method of the present invention.
Fig. 2 is an effect diagram of the method.
Detailed Description
The method of the invention is verified in further detail below with reference to the accompanying drawings.
As shown in fig. 1, a fast shot storm signal searching method based on a target detection model includes the following steps:
step 1: a dataset is constructed.
In particular, the data set is generated by means of a code-generated fast-shot analogue signal. For example, the DM value, the signal-to-noise ratio, and the full width at half maximum are randomly generated as set ranges. And the data sets are partitioned by the training set and the validation set in a set ratio (e.g., a ratio of 9: 1).
In addition, a batch of analog data (e.g., 880 sheets) is regenerated as a test set.
Step 2: the Mask R-CNN model was trained with simulation data.
Firstly, setting parameters to obtain the best effect.
Then, a backbone feature extraction network Resnet101 and a feature pyramid network FPN (Feature Pyramid Networks) are used as feature extraction layers.
For example, the feature map of Conv2_1 (Convolt2_1, the first Convolution layer inside the second Convolution block), conv3_2 (Convolt3_2, the second Convolution layer inside the third Convolution block), conv4_3 (Convolt4_3, the third Convolution layer inside the fourth Convolution block), conv5_4 (Convolt5_4, the fourth Convolution layer inside the fifth Convolution block) of ResNet is taken out and put into the FPN for operation.
The FPN takes out the feature images of each layer of the original trunk network and then operates the feature images, so that the problem that small-size objects in an object detection scene are difficult to detect can be effectively solved.
FPN takes the form of a Feature Map within a pyramid like SSD (Single Shot MultiBox Detector). Unlike SSD, FPN uses not only Feature maps of deep layer in VGG (Visual Geometry Group), but also Feature maps of shallow layer are applied to FPN. By means of bottom-up (bottom-up), top-down (top-down) and transverse connection (lateral connection), the Feature maps are integrated efficiently, the detection time is not greatly increased while the accuracy is improved, SSD is used as a target detection algorithm, and is one of main detection frames up to the present; VGG is a convolutional neural network structure that is widely used; resnet101 represents a backbone feature extraction network, resNet101 has two basic blocks, namely Conv Block and Identity Block, wherein the input and output dimensions of Conv Block are different, so that the network cannot be continuously connected in series, the network dimension is changed, the Identity Block is the same in input dimension and output dimension, so that the network can be connected in series, and the network is deepened.
Step 3: the model is tested with simulated data.
And using the cross-over ratio IoU and the Recall ratio Recall as judging model effects, wherein:
a represents a prediction frame region, B represents a label frame region, A.u.B represents a portion where two regions overlap, and A.u.B represents a set of the two regions. The larger the cross-over ratio, the more accurate the model searches for the FRB signal location.
TP represents an FRB signal predicted as an FRB signal, FP represents a non-FRB signal predicted as an FRB signal, TN represents a non-FRB signal predicted as a non-FRB signal, and FN represents an FRB signal predicted as a non-FRB signal. The higher the recall, the lower the probability that the representative model missed the FRB signal.
Step 4: and constructing a DM value regression model, wherein the DM value regression model comprises a Resnet50 model and a full link layer.
And finally adding a full connection layer by using the Resnet50 model to adapt to the DM value problem, and sending the model into a network for training.
During training, the pictures and the characteristic labels thereof cut by the rapid radio storm search model based on target detection are written into the GPU, the GPU is used for accelerating training, and the pictures are sent into the model training, including flattening matrix, calculating mean square error, gradient zero clearing, back propagation and updating optimizer operation.
Step 5: combining the search model and the DM value regression model, taking the output of the search model as the input of the regression model, and training the model.
Step 6: the entire fast-shot basic processing model is tested.
By R 2 As the judgment model effect, the following is concrete:
wherein m represents the number of samples, y (i) The true value is represented by a value that is true,representing predicted values +.>Representing the sample mean. Var (y) represents the mean-generated error, < >>Representing the prediction-induced error. R is R 2 When=1, it means that the predicted value and the true value in the sample are completely equal, without any error. The best-performing model, meaning R 2 The value also reaches a maximum.
Step 7: data, which may contain a fast-shot signal, is input into the model, and signal location information and DM values are output.
Specifically, data captured by the radio telescope, which may contain a rapid radio storm signal, is first input into a rapid radio storm base processing model. The data possibly containing the rapid radio storm signal is firstly passed through the RPN network of the search model to generate a candidate region, a binding box (boundary box) is obtained, the data containing the rapid radio storm signal is detected, and the search model integrates the data containing the rapid radio storm signal and is used as the input of the regression model. And after the regression model is calculated, outputting DM values of all the rapid radio storm data in the data.
The method introduces a target detection technology to search the rapid electric-radiation storm signal instead of manual searching, and searches the rapid electric-radiation storm signal without dispersion, thereby improving the searching efficiency of the rapid electric-radiation signal. The DM value regression model is introduced to replace the traditional method for violently cracking the DM value, so that the achromatizing efficiency is improved, and the searching efficiency of the rapid radio storm signal is indirectly improved, as shown in figure 2. The method has wide application prospect in the field of astronomical signal search.
The foregoing is a preferred embodiment of the invention and the invention should not be limited to this embodiment and the disclosure of the drawings. All equivalents and modifications that come within the spirit of the disclosure are desired to be protected.
Claims (5)
1. A rapid radio storm signal searching method based on a target detection model is characterized by comprising the following steps:
step 1: constructing a data set;
step 2: training a Mask R-CNN model by using simulation data;
firstly, setting parameters to obtain the best effect;
then, using a backbone feature extraction network Resnet101 and a feature pyramid network FPN as feature extraction layers;
the FPN takes out each layer of side feature images of the original trunk network and then operates the feature images;
the FPN adopts a form of Feature Map in the pyramid; the FPN not only uses the Feature Map with deep hierarchy in the convolutional neural network structure VGG, but also is applied to the FPN, and the Feature Map is integrated through bottom-up, top-down and transverse connection;
resnet101 represents a backbone feature extraction network, resNet101 has two basic blocks, namely Conv Block and Identity Block, wherein the input and output dimensions of Conv Block are different, the effect is to change the dimension of the network, and the input dimension and output dimension of Identity Block are the same, the effect is to deepen the network;
step 3: testing the model with simulated data;
and using the cross-over ratio IoU and the Recall ratio Recall as judging model effects, wherein:
wherein A represents a predicted frame region, B represents a marked frame region, A n B represents a part where two regions overlap, A n B represents a set of the two regions, and the larger the intersection ratio is, the more accurate the model searches FRB signal positions are:
where TP represents an FRB signal predicted as an FRB signal, FP represents a non-FRB signal predicted as an FRB signal, TN represents a non-FRB signal predicted as a non-FRB signal, and FN represents an FRB signal predicted as a non-FRB signal; the higher the recall rate, the lower the probability of missing FRB signals of the representative model;
step 4: constructing a DM value regression model, including a Resnet50 model and a full link layer;
using a Resnet50 model, and finally adding a full connection layer to adapt to the problem of DM values, and sending the model into a network for training;
during training, writing the pictures and the characteristic labels thereof cut by the rapid radio storm search model based on target detection into a GPU, using the GPU to accelerate training, and sending the pictures into the model training, wherein the operations comprise flattening matrix, calculating mean square error, gradient zero clearing, back propagation and updating an optimizer;
step 5: combining the search model and the DM value regression model, taking the output of the search model as the input of the regression model, and training the model;
step 6: testing the whole rapid radio storm basic processing model;
step 7: data, which may contain a fast-shot signal, is input into the model, and signal location information and DM values are output.
2. The method of claim 1, wherein step 1 comprises:
generating a data set by adopting a mode of generating a rapid radio storm analog signal by means of codes, and dividing the data set according to a training set and a verification set in a set proportion; in addition, a batch of simulation data is regenerated as a test set.
3. The method for searching fast radio storm signals based on object detection model as claimed in claim 2, wherein in step 1.2, the DM value, signal to noise ratio, full width at half maximum are randomly generated into analog signals according to a set range; the training set and the validation set are partitioned into data sets at a 9:1 ratio.
4. The method of claim 1, wherein in step 6, R is used to search for a fast-speed radio storm signal based on a target detection model 2 As a judgment model effect:
wherein m represents the number of samples, y (i) The true value is represented by a value that is true,representing predicted values +.>Representing a sample mean; var (y) represents the mean-generated error, < >>Representing the prediction-generated error; r is R 2 When=1, it means that the predicted value and the true value in the sample are completely equal, without any error.
5. The method of claim 1, wherein in step 7, data captured by the radio telescope and possibly containing the rapid-response storm signal is input into a rapid-response storm basic processing model;
the data possibly containing the rapid radio storm signals firstly pass through an RPN network of a search model to generate candidate areas, a Bounding box is obtained, the data containing the rapid radio storm signals are detected, and then the data containing the rapid radio storm signals are integrated by the search model to be used as the input of a regression model;
and after the regression model is calculated, outputting DM values of all the rapid radio storm data in the data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311024924.0A CN117054754A (en) | 2023-08-15 | 2023-08-15 | Quick radio storm signal searching method based on target detection model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311024924.0A CN117054754A (en) | 2023-08-15 | 2023-08-15 | Quick radio storm signal searching method based on target detection model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117054754A true CN117054754A (en) | 2023-11-14 |
Family
ID=88660344
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311024924.0A Pending CN117054754A (en) | 2023-08-15 | 2023-08-15 | Quick radio storm signal searching method based on target detection model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117054754A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117556266A (en) * | 2024-01-11 | 2024-02-13 | 之江实验室 | Model training method, signal identification method and device |
-
2023
- 2023-08-15 CN CN202311024924.0A patent/CN117054754A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117556266A (en) * | 2024-01-11 | 2024-02-13 | 之江实验室 | Model training method, signal identification method and device |
| CN117556266B (en) * | 2024-01-11 | 2024-03-22 | 之江实验室 | Signal identification model training method, signal identification method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112731309B (en) | Active interference identification method based on bilinear efficient neural network | |
| CN110097053B (en) | Improved fast-RCNN-based electric power equipment appearance defect detection method | |
| CN110222701B (en) | Automatic bridge disease identification method | |
| CN108229550B (en) | Cloud picture classification method based on multi-granularity cascade forest network | |
| CN107563433B (en) | Infrared small target detection method based on convolutional neural network | |
| Tan et al. | CALPA-NET: Channel-pruning-assisted deep residual network for steganalysis of digital images | |
| CN113569667A (en) | Inland ship target identification method and system based on lightweight neural network model | |
| CN112613350A (en) | High-resolution optical remote sensing image airplane target detection method based on deep neural network | |
| CN116206185A (en) | Lightweight small target detection method based on improved YOLOv7 | |
| CN117054754A (en) | Quick radio storm signal searching method based on target detection model | |
| CN113792685B (en) | Microseism event detection method based on multi-scale convolutional neural network | |
| CN110599459A (en) | Underground pipe network risk assessment cloud system based on deep learning | |
| CN115311502A (en) | Remote sensing image small sample scene classification method based on multi-scale double-flow architecture | |
| CN114155474A (en) | Damage identification technology based on video semantic segmentation algorithm | |
| CN111783616B (en) | Nondestructive testing method based on data-driven self-learning | |
| CN110599458A (en) | Underground pipe network detection and evaluation cloud system based on convolutional neural network | |
| CN115272278A (en) | Method for constructing change detection model for remote sensing image change detection | |
| CN116699096A (en) | Water quality detection method and system based on deep learning | |
| CN111815526A (en) | Rain image rainstrip removing method and system based on image filtering and CNN | |
| CN114299305A (en) | Salient object detection algorithm for aggregating dense and attention multi-scale features | |
| CN117218545A (en) | LBP feature and improved Yolov 5-based radar image detection method | |
| CN117152094A (en) | Method, device and system for analyzing surface defects of steel plate based on computer vision | |
| CN110728316A (en) | Classroom behavior detection method, system, device and storage medium | |
| CN110599460A (en) | Underground pipe network detection and evaluation cloud system based on hybrid convolutional neural network | |
| CN115375654A (en) | Fluorescent logging system based on machine vision and detection method thereof |
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 |