CN112070721A - Antenna parameter measuring method and device based on instance segmentation network and storage medium - Google Patents
Antenna parameter measuring method and device based on instance segmentation network and storage medium Download PDFInfo
- Publication number
- CN112070721A CN112070721A CN202010811969.2A CN202010811969A CN112070721A CN 112070721 A CN112070721 A CN 112070721A CN 202010811969 A CN202010811969 A CN 202010811969A CN 112070721 A CN112070721 A CN 112070721A
- Authority
- CN
- China
- Prior art keywords
- antenna
- network
- inputting
- module
- image information
- 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.)
- Granted
Links
- 230000011218 segmentation Effects 0.000 title claims abstract description 67
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000005259 measurement Methods 0.000 claims abstract description 38
- 238000000605 extraction Methods 0.000 claims abstract description 25
- 238000001914 filtration Methods 0.000 claims abstract description 11
- 230000007246 mechanism Effects 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 17
- 238000013526 transfer learning Methods 0.000 claims description 13
- 238000000691 measurement method Methods 0.000 claims description 11
- 238000005070 sampling Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 9
- 238000012512 characterization method Methods 0.000 claims description 6
- 230000009977 dual effect Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 7
- 238000011176 pooling Methods 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000000873 masking effect Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000010295 mobile communication Methods 0.000 description 2
- 238000005192 partition Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000012938 design process Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Quality & Reliability (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an antenna parameter measuring method, device and storage medium based on an example division network, comprising the steps of obtaining antenna pattern information; inputting the antenna image information into a network for feature extraction to obtain a first feature quantity; inputting the first characteristic quantity into a double-attention machine module to obtain an accurate boundary frame; inputting the first characteristic quantity into a characteristic pyramid network to obtain a prototype mask graph, inputting the prototype mask graph into a PointRend module, and combining an accurate bounding box and threshold filtering to obtain a smooth instance segmentation mask graph; and performing data fitting on the smooth instance segmentation mask map to obtain antenna parameters, thereby realizing safe, rapid and high-precision antenna parameter measurement.
Description
Technical Field
The invention relates to the field of antenna inclination angle measurement, in particular to an antenna parameter measurement method and device based on an example segmentation network and a storage medium.
Background
The antenna downward inclination angle is the angle between the antenna on the signal tower and the vertical direction, and also becomes the mechanical downward inclination angle of the antenna. The downward inclination angle of the antenna is generally determined by comprehensively analyzing the coverage area, the terrain, the station location distribution condition of the area where the antenna is located, the hanging height, the station distance and the population density of the area where the antenna is located and other wireless propagation environments. With the change of factors such as urban population density, environment and social development, such as network optimization, the downtilt angle of the antenna needs to be detected in time to determine whether adjustment is needed. With the gradual improvement of network construction, the antenna downtilt may need to be optimized and adjusted in later-stage network optimization. At the moment, the antenna mechanical downward inclination angle information is very important, the acquisition of the antenna downward inclination angle information is often difficult for network optimization personnel, and the traditional measurement method needs a professional tower worker to go up the tower for measurement, which consumes manpower and time, and is difficult to ensure safety.
Disclosure of Invention
The present invention is directed to solving at least one of the problems of the prior art.
Therefore, the invention provides an antenna parameter measurement method based on an example segmentation network, which can realize safe, rapid and high-precision antenna parameter measurement.
The invention also provides an antenna parameter measuring device based on the example segmentation network, which applies the antenna parameter measuring method based on the example segmentation network.
The invention also provides a computer readable storage medium applying the antenna parameter measuring method based on the example segmentation network.
The antenna parameter measuring method based on the example partition network according to the embodiment of the first aspect of the invention comprises the following steps:
acquiring antenna pattern information;
inputting the antenna image information into a network for feature extraction to obtain a first feature quantity;
inputting the first characteristic quantity into a double-attention machine module to obtain an accurate boundary frame;
inputting the first characteristic quantity into a characteristic pyramid network to obtain a prototype mask graph;
inputting the prototype mask graph into a PointRend module, and combining the accurate bounding box and threshold filtering to obtain a smooth instance segmentation mask graph;
and performing data fitting on the smooth example segmentation mask map to obtain antenna parameters.
The antenna parameter measuring method based on the example segmentation network provided by the embodiment of the invention at least has the following beneficial effects: firstly, antenna image information is obtained, and then the antenna image information is subjected to feature extraction by using a network; the problems of high training difficulty and overfitting caused by the problem of small samples can be well solved by transfer learning; the double attention mechanism module can well improve the accuracy of the network; the edges of the prototype mask graph in the pyramid network can be well smoothed by the PointRend module, and the smooth example mask graph is fitted by a data fitting method, so that antenna parameters are obtained, rapid and high-precision antenna parameter measurement is realized, manual measurement is not needed, and the antenna parameter measurement is labor-saving and safer.
According to some embodiments of the invention, the obtaining antenna image information comprises:
and controlling a camera to shoot and collecting the antenna image information.
According to some embodiments of the present invention, inputting the antenna image information into a network for feature extraction to obtain a first feature quantity includes:
acquiring the weight of a pre-training model through transfer learning;
training a network by loading the weights;
and inputting the antenna image information into the network for feature extraction to obtain a first feature quantity.
According to some embodiments of the invention, the dual attention mechanism module comprises a channel attention mechanism module and a spatial attention mechanism module.
According to some embodiments of the invention, the inputting the first feature quantity to a dual-attention mechanism module to obtain an accurate bounding box comprises:
inputting the first characteristic quantity into the channel attention module to obtain an intermediate characteristic quantity;
and inputting the intermediate characteristic quantity into the space attention module to obtain an accurate bounding box.
According to some embodiments of the invention, the inputting the prototype mask graph to a PointRend module, in combination with the exact bounding box and threshold filtering, results in a smooth instance segmentation mask graph, comprising:
selecting a plurality of sampling points from the prototype mask graph;
extracting the characteristics of the sampling points to obtain a characterization vector;
and predicting the characterization vector by using a multilayer perceptron, and combining the precise bounding box and threshold filtering to obtain a smooth instance segmentation mask map.
According to some embodiments of the present invention, the performing data fitting on the smoothed instance segmentation mask map to obtain antenna parameters includes:
and performing data fitting on the smooth instance segmentation mask map by using a least square method to obtain antenna parameters.
The antenna parameter measuring device of the example-based segmentation network according to the second aspect of the present invention includes:
the acquisition unit is used for acquiring antenna image information;
the extraction unit is used for inputting the antenna image information into a network for feature extraction to obtain a first feature quantity;
the first processing unit is used for inputting the first characteristic quantity into the double-attention machine module to obtain an accurate boundary frame;
the second processing unit is used for inputting the first characteristic quantity into the characteristic pyramid network to obtain a prototype mask graph; the segmentation unit is used for inputting the prototype mask graph into a PointRend module, and combining the accurate bounding box and threshold filtering to obtain a smooth instance segmentation mask graph;
and the calculation unit is used for performing data fitting on the smooth instance segmentation mask map to obtain antenna parameters.
According to some embodiments of the invention, the first processing unit comprises:
the first operation unit is used for inputting the first characteristic quantity into the channel attention module to obtain an intermediate characteristic quantity;
and the second operation unit is used for inputting the intermediate characteristic quantity into the space attention module to obtain an accurate bounding box.
The antenna parameter measuring device based on the example segmentation network provided by the embodiment of the invention at least has the following beneficial effects: firstly, antenna image information is obtained, and then the antenna image information is subjected to feature extraction by using a network; the problems of high training difficulty and overfitting caused by the problem of small samples can be well solved by transfer learning; the double attention mechanism module can well improve the accuracy of the network; the edges of the prototype mask graph in the pyramid network can be well smoothed by the PointRend module, and the smooth example mask graph is fitted by a data fitting method, so that antenna parameters are obtained, rapid and high-precision antenna parameter measurement is realized, manual measurement is not needed, and the antenna parameter measurement is labor-saving and safer.
According to the computer-readable storage medium of the third aspect of the present invention, the antenna parameter measurement method of the example-based segmentation network according to the first aspect of the present invention can be applied.
The computer-readable storage medium according to the embodiment of the invention has at least the following advantages: firstly, antenna image information is obtained, and then the antenna image information is subjected to feature extraction by using a network; the problems of high training difficulty and overfitting caused by the problem of small samples can be well solved by transfer learning; the double attention mechanism module can well improve the accuracy of the network; the edges of the prototype mask graph in the pyramid network can be well smoothed by the PointRend module, and the smooth example mask graph is fitted by a data fitting method, so that antenna parameters are obtained, rapid and high-precision antenna parameter measurement is realized, manual measurement is not needed, and the antenna parameter measurement is labor-saving and safer.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The above and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
fig. 1 is a flowchart of an antenna parameter measurement method based on an example segmentation network according to a first embodiment of the present invention;
fig. 2 is a flowchart illustrating feature extraction in an antenna parameter measurement method based on an example segmentation network according to a first embodiment of the present invention;
fig. 3 is a flowchart illustrating a work flow of determining a bounding box in an antenna parameter measurement method based on an example segmentation network according to a first embodiment of the present invention;
fig. 4 is a flowchart illustrating example division in an antenna parameter measurement method based on an example division network according to a first embodiment of the present invention;
fig. 5 is a schematic structural diagram of an antenna parameter measuring device based on an example segmentation network according to a second embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.
In the description of the present invention, unless otherwise explicitly defined, terms such as arrangement, connection and the like should be broadly construed, and those skilled in the art can reasonably determine the specific meanings of the above terms in the present invention in combination with the detailed contents of the technical solutions.
Example one
Referring to fig. 1, an embodiment of the present invention provides an antenna parameter measurement method based on an example-partitioned network, where one embodiment includes, but is not limited to, the following steps:
step S100, antenna image information is acquired.
In this embodiment, the step of acquiring the antenna pattern information is prepared for the measurement of the antenna parameters, and is the basis of the measurement of the antenna parameters.
Step S200, inputting the antenna image information into a network for feature extraction to obtain a first feature quantity.
In this embodiment, in this step, antenna image information is input into a network for feature extraction, so as to obtain a first feature quantity; the pre-training model can well overcome the problems of overfitting, difficulty in training and the like caused by less sample data.
And step S300, inputting the first characteristic quantity into a double-attention machine module to obtain an accurate boundary frame.
In this embodiment, the first feature quantity is input into the dual-attention machine module in this step, so as to obtain the accurate bounding box, where the dual-attention machine module can well improve the example segmentation accuracy of the network.
Step S400, inputting the first characteristic quantity into the characteristic pyramid network to obtain a prototype mask image.
In this embodiment, in this step, the first feature quantity is input into the feature pyramid network to obtain a prototype mask pattern, which is prepared for antenna parameter measurement.
Step S500, inputting the prototype mask graph into a PointRend module, and combining an accurate bounding box and threshold filtering to obtain a smooth instance segmentation mask graph.
In this embodiment, the masking map is input to the PointRend module in this step, so that the edge of the masking map can be smoothed, and the segmentation accuracy can be improved well.
And step S600, performing data fitting on the smooth instance segmentation mask map to obtain antenna parameters.
In this embodiment, in this step, data fitting is performed on the smooth instance segmentation mask map, and antenna parameters are calculated, so that the method is simple, convenient and fast.
In step S100 of this embodiment, the following steps may be included, but not limited to:
and controlling a camera to shoot and collecting the antenna image information.
In this embodiment, in this step, the camera is controlled to shoot, for example, by setting the camera on the unmanned aerial vehicle, the antenna is shot, the antenna image information is collected, and a precondition is prepared for measuring the antenna parameters.
Referring to fig. 2, in step S200 of this embodiment, the following steps may be included, but are not limited to:
step S210, acquiring the weight of the pre-training model through transfer learning.
In this embodiment, the weight parameters of the pre-training model are obtained through the transfer learning in this step, and the transfer learning method can well alleviate the problems of high training difficulty, overfitting and the like caused by the problem of small samples.
Step S220, train the network by loading the weights.
In this embodiment, this step trains the network by loading the weights, and prepares for feature extraction.
Step S230, inputting the antenna image information into the network for feature extraction, so as to obtain a first feature quantity.
In this embodiment, in this step, antenna image information is input into a network for feature extraction, so as to obtain a first feature quantity, and a precondition is prepared for measurement of antenna parameters.
In some embodiments of the invention, the dual attention mechanism module includes a channel attention mechanism module and a spatial attention mechanism module. The channel attention mechanism module uses maximum pooling and average pooling output through a shared network; the spatial attention mechanism module combines two similar outputs along the channel axis and maps them to the convolutional layer.
Referring to fig. 3, in step S300 of this embodiment, the following steps may be included, but are not limited to:
step S310, inputting the first characteristic quantity into a channel attention module to obtain an intermediate characteristic quantity.
In this embodiment, the step inputs the first feature quantity into the channel attention module, so as to obtain an intermediate feature quantity, which is ready for subsequent network segmentation.
Step S320, inputting the intermediate feature quantity into the spatial attention module to obtain an accurate bounding box.
In the present embodiment, this step inputs the intermediate feature quantity into the spatial attention module, so that an accurate bounding box can be obtained.
Referring to fig. 4, in step S500 of the present embodiment, the following steps may be included, but not limited to:
step S510, selecting a plurality of sampling points from the prototype mask map.
In this embodiment, this step selects a number of sampling points from the mask map, particularly points at edge positions in the mask map.
And step S520, performing feature extraction on the sampling points to obtain a characterization vector.
In this embodiment, in this step, feature extraction is performed on the selected sampling points, so as to obtain a characterization vector.
And step S530, predicting the eigenvector by using a multilayer perceptron, and combining an accurate boundary box and threshold filtering to obtain a smooth instance segmentation mask map.
In this embodiment, in this step, the multilayer perceptron is used to predict the eigenvector, so that a smooth instance segmentation mask map can be obtained, the edges of the prototype mask map are well smoothed, and the accuracy of network segmentation is improved.
In step S600 of this embodiment, the following steps may be included, but not limited to:
and performing data fitting on the smooth instance segmentation mask map by using a least square method to obtain antenna parameters.
In this embodiment, in this step, a least square method is used to perform linear fitting, and the antenna parameters are obtained from the smoothed instance segmentation mask map, so that the measurement of the antenna parameters can be more accurate.
A specific embodiment is provided below to further explain the antenna parameter measurement method based on the example split network:
the embodiment of the invention mainly provides an antenna parameter measuring method based on an example segmentation network, which combines an attention mechanism and a PointRend module to improve the segmentation precision and adopts a single-step example segmentation network to improve the segmentation speed. Wherein, the RetinaNet based on the feature pyramid network and the rescet 101 is used to construct a basic network framework, and a double attention mechanism module is used to perform channel and space attention on each convolution block of the deep network to complete the feature map of the middle layer, and a PointRend module is added after the mask map to generate a finer instance mask.
Because the data with the labels are difficult to obtain, the labeled data are difficult, the small samples have the problems of overfitting and the like, the embodiment of the invention trains the antenna segmentation network by obtaining parameters from the large database model trained by the deep convolutional neural network by using the transfer learning, so as to achieve the purpose of high-precision segmentation.
The double-attention machine module can be integrated into any convolutional neural network architecture seamlessly due to the fact that the design process of the double-attention machine module is very light, extra overhead can be ignored, and meanwhile end-to-end training can be conducted with any basic CNN framework. The dual attention machine is made with two successive sub-modules: channel and space sub-modules. Wherein the channel attention mechanism module uses maximum pooling and average pooling outputs over a shared network; the spatial attention mechanism module combines two similar outputs along the channel axis and maps them to convolutional layers. In the channel attention module, each channel in the feature map represents a detector, so it is said that the features of interest of the channel attention mechanism are meaningful. To collect spatial features, a global average pool and a maximum pool are used to obtain different feature information, which may be described as follows:
the characteristic F of H × W × C is input. Spatial global average pooling and maximum pooling are first performed separately to obtain two 1 × 1 × C channel information. Then, the neurons are respectively sent to a two-layer neural network, the number of the neurons in the first layer is C/r, the activation function is ReLU, and the number of the neurons in the second layer is C. The two-layer neural network is feature-shared. Then, the obtained two features are added and the function is activated by sigmoid to obtain a weight coefficient. And finally, multiplying the weight coefficient by the original characteristic F to obtain a new scaled characteristic.
After passing through the attention mechanism module, a spatial attention module, which will be described in detail below, is used where the attention features are meaningful.
The spatial attention mechanism is similar to the channel attention mechanism, with feature F being H × W × C, first, the channel dimensions are averaged pooled and maximal pooled to obtain two H × W × 1 channel information, and the two information are combined together into a channel. Subsequently, after 7 × 7 convolutional layers and sigmoid activation functions, weights can be obtainedCoefficient Ms. Finally, the weight coefficient M is calculatedsMultiplying the new feature by the feature F to obtain a new scaled feature. Both the channel attention and spatial attention modules can be combined in parallel or sequentially with any basic CNN, and experiments have shown that sequential combining and placing the channel attention mechanism first can achieve better results. The channel attention module and the space attention module are used as two independent parts, so that parameters and computing power can be saved, and the channel attention module and the space attention module can be integrated into an existing network architecture as a plug-and-play module.
Due to the over-sampling and under-sampling problems in the pixel labeling task, the mask branch cannot add a very fine and precise mask for each situation, which will greatly affect the measurement accuracy of the antenna parameter measurement. Thus, an efficient and high quality PointRend module is combined to smooth the edges of the segmented instance mask and minimize measurement errors of the antenna parameters. PointRend is mainly composed of three parts: point selection strategy, point-like feature representation and point prediction. The implementation of the point selection strategy mainly depends on a flexible adaptive selection of suitable sampling points. For example, more points are sampled in the boundary region and fewer points are sampled in the non-boundary region. After selecting the point strategy, the most uncertain points are selected to re-predict the segmentation results of these points. After the feature of each special point (most uncertain point) is calculated by the point-like feature representation module, the feature of each point is processed by point prediction to obtain a new prediction result. The above steps can be summarized as special treatment of points at boundary positions to make the result more accurate. Therefore, the mask edge of the mobile communication base station antenna can be finely smoothed by a point to improve the fitting accuracy and the parameter measurement accuracy.
A downtilt fitting structure with a PointRend module. The least squares method is the most basic and important data fitting method for revealing the relationship between pixels in the mask to quantify the antenna parameters. The least squares method has wide application in data modeling. The utilization of this theory to solve the problem has a very important role and position, especially in the research of mass data analysis. To reduce measurement error and computational cost, the edge 3/5 pixel value coordinates on both sides of the mask (top and bottom of mask are removed 1/5) are selected to accommodate the downtilt angle FIG. 5 shows the PointRend module function with a parameter fitting structure.
First, assume that the parameter estimation equation is as follows, where x and y denote an abscissa point and an ordinate point of a pixel value coordinate point selected in a mask, respectively, β denotes a correlation coefficient, and μ denotes a deviation.
y=β0+β1x+β2x+…+βkx+μ
Substituting n sets of pixel value coordinates into the above equation yields:
the matrix representation of the above formula is given by:
y=BX+μ
thus, the sum of the squares of the differences between the observed values and the estimated values of the explained variables is:
obtaining the following components according to the principle of least square method:
thus, the least squares estimate of the parameters is:
among them, the deep convolutional neural network has the ability to learn rich image representation, and is therefore widely used in various large application scenarios. However, to learn these functions, a large amount of data is required. For the data sets used in the present invention, the amount of data is insufficient, resulting in overfitting and training difficulties. Transfer learning is a method of transferring knowledge from the relevant domain to a new problem. The migration learning strategy learns low-level and mid-level features from the migrated domain, and therefore higher performance can be achieved by taking a small amount of data from the new domain. Therefore, applying the transfer learning will be trained on massive amounts
The recognition model on the image is transferred to the task of mobile communication base station antenna segmentation, so that the trouble caused by small samples can be effectively reduced, and the performance same as that of the task of mass data can be obtained.
According to the scheme, antenna image information is obtained firstly, and then the antenna image information is subjected to feature extraction by using a network; the problems of high training difficulty and overfitting caused by the problem of small samples can be well solved by transfer learning; the double attention mechanism module can well improve the accuracy of the network; the edges of the prototype mask graph in the pyramid network can be well smoothed by the PointRend module, and the smooth instance segmentation mask graph is fitted by a data fitting method, so that antenna parameters are obtained, rapid and high-precision antenna parameter measurement is realized, manual measurement is not needed, and the antenna parameter measurement is labor-saving and safer.
Example two
Referring to fig. 5, a second embodiment of the present invention provides an antenna parameter measurement apparatus 1000 based on an example partition network, including:
an acquisition unit 1100, configured to acquire antenna pattern information;
an extracting unit 1200, configured to input the antenna image information into a pre-training model for feature extraction, so as to obtain a first feature quantity;
the first processing unit 1300 is configured to input the first feature quantity to the dual-attention machine module to obtain an accurate bounding box;
the second processing unit 1400 is configured to input the first feature quantity to the feature pyramid network to obtain a prototype mask map;
a segmentation unit 1500, configured to input the prototype mask map into a PointRend module, so as to obtain a smooth instance segmentation mask map;
a calculating unit 1600, configured to perform data fitting on the smoothed instance segmentation mask map to obtain an antenna parameter.
In this embodiment, the first processing unit 1300 includes:
a first operation unit 1310 configured to input the first feature amount into the channel attention module to obtain an intermediate feature amount;
a second operation unit 1320, configured to input the intermediate feature quantity into the spatial attention module, so as to obtain an accurate bounding box.
It should be noted that, since the antenna parameter measuring apparatus based on the example division network in the present embodiment is based on the same inventive concept as the antenna parameter measuring method based on the example division network in the first embodiment, the corresponding content in the first method embodiment is also applicable to the present system embodiment, and is not described in detail herein.
According to the scheme, antenna image information is obtained firstly, and then the antenna image information is subjected to feature extraction by using a network; the problems of high training difficulty and overfitting caused by the problem of small samples can be well solved by transfer learning; the double attention mechanism module can well improve the accuracy of the network; the edges of the mask images which are segmented by the double-attention machine model can be well smoothed by the PointRend module, and the smooth instance segmentation mask images are fitted by a data fitting method, so that antenna parameters are obtained, rapid and high-precision antenna parameter measurement is realized, manual measurement is not needed, and the antenna parameter measurement is labor-saving and safer.
EXAMPLE III
A third embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium stores an executable instruction of an antenna parameter measurement device based on an example segmentation network, where the executable instruction of the antenna parameter measurement device based on the example segmentation network is used to enable the antenna parameter measurement device based on the example segmentation network to perform the above-mentioned method for measuring antenna parameters based on the example segmentation network, for example, to perform the above-described method steps S100 to S600 in fig. 1, and implement the function of the unit 1000-1500 in fig. 5.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an illustrative embodiment," "an example," "a specific example," or "some examples" or the like mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
Claims (10)
1. The method for measuring the antenna parameters based on the example segmentation network is characterized by comprising the following steps:
acquiring antenna pattern information;
inputting the antenna image information into a network for feature extraction to obtain a first feature quantity;
inputting the first characteristic quantity into a double-attention machine module to obtain an accurate boundary frame;
inputting the first characteristic quantity into a characteristic pyramid network to obtain a prototype mask graph;
inputting the prototype mask graph into a PointRend module, and combining the accurate bounding box and threshold filtering to obtain a smooth instance segmentation mask graph;
and performing data fitting on the smooth example segmentation mask map to obtain antenna parameters.
2. The method for measuring antenna parameters based on the example split network according to claim 1, wherein the obtaining of the antenna image information comprises:
and controlling a camera to shoot and collecting the antenna image information.
3. The method for measuring antenna parameters based on the example segmentation network according to claim 1, wherein the inputting the antenna image information into the network for feature extraction to obtain a first feature quantity comprises: acquiring the weight of a pre-training model through transfer learning;
training a network by loading the weights;
and inputting the antenna image information into the network for feature extraction to obtain a first feature quantity.
4. The method of claim 1, wherein the method comprises:
the dual attention mechanism module includes a channel attention mechanism module and a spatial attention mechanism module.
5. The method for antenna parameter measurement based on the example segmentation network according to claim 4, wherein the inputting the first feature quantity into a dual attention mechanism module to obtain an accurate bounding box comprises:
inputting the first characteristic quantity into the channel attention module to obtain an intermediate characteristic quantity;
and inputting the intermediate characteristic quantity into the space attention module to obtain an accurate bounding box.
6. The method for antenna parameter measurement based on the example segmentation network of claim 1, wherein the inputting the prototype mask map to a PointRend module, and combining the precise bounding box and threshold filtering to obtain a smoothed example segmentation mask map comprises:
selecting a plurality of sampling points from the prototype mask graph;
extracting the characteristics of the sampling points to obtain a characterization vector;
and predicting the characterization vector by using a multilayer perceptron, and combining the accurate bounding box and threshold filtering to obtain a smooth instance segmentation mask map.
7. The method according to claim 1, wherein the step of performing data fitting on the smoothed instance segmentation mask map to obtain the antenna parameters comprises:
and performing data fitting on the smooth instance segmentation mask map by using a least square method to obtain antenna parameters.
8. An antenna parameter measurement device for an instance based split network, comprising:
the acquisition unit is used for acquiring antenna image information;
the extraction unit is used for inputting the antenna image information into a network for feature extraction to obtain a first feature quantity;
the first processing unit is used for inputting the first characteristic quantity into the double-attention machine module to obtain an accurate boundary frame;
the second processing unit is used for inputting the first characteristic quantity into the characteristic pyramid network to obtain a prototype mask graph; the segmentation unit is used for inputting the prototype mask graph into a PointRend module, and combining the accurate bounding box and threshold filtering to obtain a smooth instance segmentation mask graph;
and the calculation unit is used for performing data fitting on the smooth instance segmentation mask map to obtain antenna parameters.
9. The apparatus of claim 8, wherein the first processing unit comprises:
the first operation unit is used for inputting the first characteristic quantity into the channel attention module to obtain an intermediate characteristic quantity;
and the second operation unit is used for inputting the intermediate characteristic quantity into the space attention module to obtain an accurate bounding box.
10. A computer-readable storage medium characterized by: the computer-readable storage medium stores example-split-network-based antenna parameter measurement device-executable instructions for causing an example-split-network-based antenna parameter measurement device to perform the example-split-network-based antenna parameter measurement method according to any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010811969.2A CN112070721B (en) | 2020-08-13 | 2020-08-13 | Antenna parameter measurement method, device and storage medium based on instance division network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010811969.2A CN112070721B (en) | 2020-08-13 | 2020-08-13 | Antenna parameter measurement method, device and storage medium based on instance division network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112070721A true CN112070721A (en) | 2020-12-11 |
CN112070721B CN112070721B (en) | 2024-01-12 |
Family
ID=73661585
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010811969.2A Active CN112070721B (en) | 2020-08-13 | 2020-08-13 | Antenna parameter measurement method, device and storage medium based on instance division network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112070721B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109458980A (en) * | 2018-11-06 | 2019-03-12 | 五邑大学 | A kind of Downtilt measurement method based on linear regression fit |
CN110532955A (en) * | 2019-08-30 | 2019-12-03 | 中国科学院宁波材料技术与工程研究所 | Example dividing method and device based on feature attention and son up-sampling |
CN111062466A (en) * | 2019-12-11 | 2020-04-24 | 南京华苏科技有限公司 | Method for predicting field intensity distribution of cell after antenna adjustment based on parameters and neural network |
CN111445474A (en) * | 2020-05-25 | 2020-07-24 | 南京信息工程大学 | Kidney CT image segmentation method based on bidirectional complex attention depth network |
-
2020
- 2020-08-13 CN CN202010811969.2A patent/CN112070721B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109458980A (en) * | 2018-11-06 | 2019-03-12 | 五邑大学 | A kind of Downtilt measurement method based on linear regression fit |
CN110532955A (en) * | 2019-08-30 | 2019-12-03 | 中国科学院宁波材料技术与工程研究所 | Example dividing method and device based on feature attention and son up-sampling |
CN111062466A (en) * | 2019-12-11 | 2020-04-24 | 南京华苏科技有限公司 | Method for predicting field intensity distribution of cell after antenna adjustment based on parameters and neural network |
CN111445474A (en) * | 2020-05-25 | 2020-07-24 | 南京信息工程大学 | Kidney CT image segmentation method based on bidirectional complex attention depth network |
Non-Patent Citations (2)
Title |
---|
王国民;谷晓鹏;路广勋;: "基于数据拟合的补全模型研究", 军事通信技术, no. 01, pages 1 - 3 * |
陈林斌;孙赐恩;: "基于最小二乘法的误差分析与误差修正", 中国新通信, no. 07, pages 1 - 2 * |
Also Published As
Publication number | Publication date |
---|---|
CN112070721B (en) | 2024-01-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110796037B (en) | Satellite-borne optical remote sensing image ship target detection method based on lightweight receptive field pyramid | |
CN110298387A (en) | Incorporate the deep neural network object detection method of Pixel-level attention mechanism | |
CN109389102A (en) | The system of method for detecting lane lines and its application based on deep learning | |
CN108122247A (en) | A kind of video object detection method based on saliency and feature prior model | |
CN108520273A (en) | A kind of quick detection recognition method of dense small item based on target detection | |
CN107909084B (en) | Haze concentration prediction method based on convolution-linear regression network | |
US11238307B1 (en) | System for performing change detection within a 3D geospatial model based upon semantic change detection using deep learning and related methods | |
CN112862090A (en) | Air temperature forecasting method based on deep space-time neural network | |
CN115457006B (en) | Unmanned aerial vehicle inspection defect classification method and device based on similarity consistency self-distillation | |
CN108230354A (en) | Target following, network training method, device, electronic equipment and storage medium | |
CN111985325A (en) | Aerial small target rapid identification method in extra-high voltage environment evaluation | |
CN111104850A (en) | Remote sensing image building automatic extraction method and system based on residual error network | |
CN115690627A (en) | Method and system for detecting aerial image rotating target | |
CN115147644A (en) | Method, system, device and storage medium for training and describing image description model | |
CN111383273A (en) | High-speed rail contact net part positioning method based on improved structure reasoning network | |
CN112069997B (en) | Unmanned aerial vehicle autonomous landing target extraction method and device based on DenseHR-Net | |
CN117830594A (en) | Light-weight target detection method based on attention mechanism knowledge distillation | |
Chen et al. | BARS: a benchmark for airport runway segmentation | |
CN116912646A (en) | Multimode remote sensing image fusion classification method based on double-branch dynamic modulation network | |
CN111798516A (en) | Method for detecting running state quantity of bridge crane equipment and analyzing errors | |
CN112070721B (en) | Antenna parameter measurement method, device and storage medium based on instance division network | |
CN115661694A (en) | Intelligent detection method, system, storage medium and electronic equipment for light-weight main transformer focusing on key characteristics | |
Spasov et al. | Transferability assessment of open-source deep learning model for building detection on satellite data | |
CN115035429A (en) | Aerial photography target detection method based on composite backbone network and multiple measuring heads | |
CN115205549A (en) | SLAM method based on mutual information and semantic segmentation |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |