CN117671504A - Marine wind power identification method and system based on yolo algorithm - Google Patents
Marine wind power identification method and system based on yolo algorithm Download PDFInfo
- Publication number
- CN117671504A CN117671504A CN202311697617.9A CN202311697617A CN117671504A CN 117671504 A CN117671504 A CN 117671504A CN 202311697617 A CN202311697617 A CN 202311697617A CN 117671504 A CN117671504 A CN 117671504A
- Authority
- CN
- China
- Prior art keywords
- wind power
- offshore wind
- module
- power identification
- model
- 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 59
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 36
- 230000007246 mechanism Effects 0.000 claims abstract description 21
- 238000004519 manufacturing process Methods 0.000 claims abstract description 20
- 238000012360 testing method Methods 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 12
- 238000012795 verification Methods 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000003860 storage Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 5
- 238000012937 correction Methods 0.000 claims description 4
- 230000005855 radiation Effects 0.000 claims description 4
- 230000002401 inhibitory effect Effects 0.000 claims description 3
- 238000001514 detection method Methods 0.000 abstract description 52
- 238000010801 machine learning Methods 0.000 abstract description 2
- 238000013135 deep learning Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 10
- 238000004891 communication Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 239000000306 component Substances 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000000670 limiting effect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 208000034699 Vitreous floaters Diseases 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000010339 dilation Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000000873 masking effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000011664 signaling Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
-
- 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
- G06V10/776—Validation; Performance evaluation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Wind Motors (AREA)
Abstract
The invention discloses a marine wind power identification method and system based on yolo algorithm, and relates to the technical field of machine learning. The method comprises the following steps: collecting remote sensing images of offshore wind farms in a plurality of areas, and manufacturing an offshore wind power sample set; constructing an initial offshore wind power identification model based on a YOLOv5 algorithm, a CA attention mechanism and an RFB multi-branch convolution module; training and verifying the initial offshore wind power identification model by utilizing the offshore wind power sample set to obtain a target offshore wind power identification model; and acquiring a real-time offshore wind power image, and identifying an offshore wind power target by using a target offshore wind power identification model. The invention can accurately and stably detect the offshore wind power, and effectively improves the detection efficiency and the detection accuracy.
Description
Technical Field
The invention relates to the technical field of machine learning, in particular to a marine wind power identification method and system based on yolo algorithm.
Background
Existing deep learning detection algorithms are mainly based on two-stage target detection networks represented by R-CNN (Girshick, 2015;Girshick et al, 2014; he et, 2017; ren et al, 2017), fast R-CNN (Girshick, 2015), fast R-CNN (Ren et al, 2017), SPP-NET (He et al, 2014), etc., and one-stage target detection networks represented by YOLO (Bochkovskiy et al, 2020;Redmon et al, 2016;Redmon and Farhadi,2018,2017), SSD (w.liu et al, 2016). The two-stage network has higher detection precision, but needs a large number of data sets, higher training cost and low detection speed. The one-stage detection method directly extracts the characteristics in the network and predicts the target category and position, has high detection speed and small required data volume, and is suitable for a small sample data set.
At present, the deep learning technology has been primarily developed in offshore wind power target identification application. Hoeser et al (2022) (Hoeser and Kuenzer, 2022) construct a global offshore wind power deep learning dataset that provides a research basis for offshore target detection. Hoeser et al (2022) (Hoeser et al, 2022) detects global offshore wind power by combining the data set with a CNN model to obtain a 2016-2021 global spatial-temporal distribution data set of offshore wind power. However, the use of the offshore wind power in the north sea (europe) and the east sea (china) as the verification set is insufficient for evaluating the global identification conditions of the offshore wind power in different sea conditions and different undersea surfaces, and the experiment still has room for improvement in the aspects of lightweight deployment and the like. Even though deep neural networks have advanced in the offshore wind detection field, deep learning-based methods still face three challenges of data set availability, model generalization, and detection result robustness. First, deep learning generally requires a large amount of sample data to train the model, thereby achieving higher regional accuracy; however, the high quality training data sets available at the present time on-source are still lacking. Secondly, the offshore wind power is sparsely distributed in the sea area, the sea surface clutter background has time phase and space differences in the remote sensing image, and the factors bring great challenges to generalization of the deep learning model. Moreover, the area of the offshore wind power target object is small, and the interference of the offshore floaters or temporarily moving objects (such as ships, clouds and oil drilling platforms) makes the wind power detection algorithm time-consuming and lack of robustness.
Therefore, how to accurately and efficiently identify the offshore wind power target in the national scale becomes a problem to be solved urgently.
Disclosure of Invention
In order to overcome the problems or at least partially solve the problems, the invention provides a marine wind power identification method and a marine wind power identification system based on a yolo algorithm, which can accurately and stably detect the marine wind power and effectively improve the detection efficiency and the detection accuracy.
In order to solve the technical problems, the invention adopts the following technical scheme:
in a first aspect, the invention provides a marine wind power identification method based on yolo algorithm, comprising the following steps:
collecting remote sensing images of offshore wind farms in a plurality of areas, and manufacturing an offshore wind power sample set;
constructing an initial offshore wind power identification model based on a YOLOv5 algorithm, a CA attention mechanism and an RFB multi-branch convolution module;
training and verifying the initial offshore wind power identification model by utilizing the offshore wind power sample set to obtain a target offshore wind power identification model;
and acquiring a real-time offshore wind power image, and identifying an offshore wind power target by using a target offshore wind power identification model.
The invention creates a deep learning training data set suitable for Chinese offshore wind power target recognition, optimizes and improves based on a YOLOv5 algorithm in a YOLO algorithm, introduces a CA attention mechanism and an RFB multi-branch convolution module, builds an effective offshore wind power recognition model (YOLOv 5s-CR model), adds a coordinate attention CA module to improve the positioning capability of the model, and combines a context enhancement module RFB to increase receptive fields to reduce the false detection rate of the model under a large-scale complex sea surface background, thereby realizing quick detection and accurate positioning of Chinese offshore wind power. The offshore wind power is accurately and stably detected based on the constructed offshore wind power identification model, and the detection efficiency and the detection accuracy are effectively improved. The marine wind power identification model constructed by the method has good generalization and geographic generalization performances.
Based on the first aspect, the method for manufacturing the offshore wind power sample set further comprises the following steps:
preprocessing remote sensing images of the offshore wind farm;
performing mean value calculation on the preprocessed remote sensing image of the offshore wind farm, and optimizing the remote sensing image of the offshore wind farm;
sample marking is carried out on the optimized remote sensing image of the offshore wind farm so as to obtain an offshore wind power sample test set and an offshore wind power sample verification set;
manufacturing sample slices of an offshore wind power sample test set and an offshore wind power sample verification set through an ArcGIS Pro image processing tool, and writing position information and category information of all samples into corresponding xml files;
and selecting part of sample slice data as a training set and a testing set to form a VOC data set.
Based on the first aspect, the method for preprocessing the remote sensing image of the offshore wind farm further comprises the following steps:
updating the orbit state vector of the remote sensing image of the offshore wind farm, and updating orbit metadata by using the corrected orbit file aiming at the ground distance multi-view image GRD in the remote sensing image of the offshore wind farm;
the method comprises the steps of inhibiting low-intensity noise and invalid data of the edge of a GRD scene through a low-value pixel mask, and eliminating additional noise in sub-strips of a remote sensing image of an offshore wind farm;
and calculating the back scattering intensity by adopting sensor calibration parameters in the metadata to perform radiation calibration, and performing orthographic correction on the remote sensing image of the offshore wind farm by combining a digital elevation model.
Based on the first aspect, the marine wind power identification method based on the yolo algorithm further comprises the following steps:
and performing image enhancement processing on the images in the data set.
Based on the first aspect, the method for training and verifying the initial offshore wind power identification model by using the offshore wind power sample set further comprises the following steps:
and training and verifying the initial offshore wind power identification model by the training set and the testing set respectively.
Based on the first aspect, the method for constructing the initial offshore wind power identification model based on the YOLOv5 algorithm, the CA attention mechanism and the RFB multi-branch convolution module further comprises the following steps:
introducing a CA attention mechanism into a backlight module of the YOLOv5 algorithm to construct a CS-CA module;
introducing an RFB multi-branch convolution module into a main network of the YOLOv5 algorithm;
and constructing an initial offshore wind power identification model based on the YOLOv5 algorithm, the CS-CA module and the RFB multi-branch convolution module.
In a second aspect, the invention provides an offshore wind power identification system based on a yolo algorithm, which comprises a sample set making module, an initial model building module, a model training and verifying module and an offshore wind power identification module, wherein:
the sample set manufacturing module is used for acquiring remote sensing images of the offshore wind farm in a plurality of areas and manufacturing an offshore wind power sample set;
the initial model building module is used for building an initial offshore wind power identification model based on a YOLOv5 algorithm, a CA attention mechanism and an RFB multi-branch convolution module;
the model training and verifying module is used for training and verifying the initial offshore wind power identification model by utilizing the offshore wind power sample set so as to obtain a target offshore wind power identification model;
and the offshore wind power identification module is used for acquiring real-time offshore wind power images and identifying an offshore wind power target by utilizing the target offshore wind power identification model.
According to the system, the offshore wind power is accurately and stably detected by the constructed offshore wind power identification model through the plurality of modules such as the sample set manufacturing module, the initial model construction module, the model training and verifying module and the offshore wind power identification module, so that the detection efficiency and the detection accuracy are effectively improved. The invention creates a deep learning training data set suitable for Chinese offshore wind power target recognition, optimizes and improves based on a YOLOv5 algorithm in a YOLO algorithm, introduces a CA attention mechanism and an RFB multi-branch convolution module, builds an effective offshore wind power recognition model (YOLOv 5s-CR model), adds a coordinate attention CA module to improve the positioning capability of the model, and combines a context enhancement module RFB to increase receptive fields to reduce the false detection rate of the model under a large-scale complex sea surface background, thereby realizing quick detection and accurate positioning of Chinese offshore wind power. The offshore wind power is accurately and stably detected based on the constructed offshore wind power identification model, and the detection efficiency and the detection accuracy are effectively improved. The marine wind power identification model constructed by the method has good generalization and geographic generalization performances.
In a third aspect, the present application provides an electronic device comprising a memory for storing one or more programs; a processor; the method of any of the first aspects described above is implemented when one or more programs are executed by a processor.
In a fourth aspect, the present application provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method as in any of the first aspects described above.
The invention has at least the following advantages or beneficial effects:
1. a deep learning training data set suitable for the recognition of the offshore wind power targets in China is created, and a precise and comprehensive data set is provided for the follow-up.
2. Based on the YOLOv5 algorithm in the YOLO algorithm, a CA attention mechanism and an RFB multi-branch convolution module are introduced, an effective offshore wind power identification model (YOLOv 5s-CR model) is constructed, the coordinate attention CA module is added to improve the positioning capability of the model, the context enhancement module RFB is fused to increase the receptive field and reduce the false detection rate of the model under the large-scale complex sea surface background, and the quick detection and accurate positioning of the offshore wind power in China can be realized. The offshore wind power is accurately and stably detected based on the constructed offshore wind power identification model, and the detection efficiency and the detection accuracy are effectively improved.
3. The marine wind power identification model constructed by the method has good generalization and geographic generalization performances.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of a marine wind power identification method based on a yolo algorithm in an embodiment of the invention;
FIG. 2 is a schematic diagram of a YOLOv5s-CR model structure and flow chart in accordance with an embodiment of the present invention;
FIG. 3 is a schematic diagram of a detection result of a timing experiment of a typical offshore wind farm in a tidal flat in a certain area according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of detection results of a typical offshore wind farm in other countries around the world in an embodiment of the present invention;
FIG. 5 is a schematic diagram of a Chinese offshore wind farm and sample labeling thereof in an embodiment of the invention;
FIG. 6 is a schematic block diagram of a marine wind power identification system based on yolo algorithm according to an embodiment of the present invention;
fig. 7 is a block diagram of an electronic device according to an embodiment of the present invention.
Reference numerals illustrate: 100. a sample set making module; 200. an initial model building module; 300. model training and verifying module; 400. an offshore wind power identification module; 101. a memory; 102. a processor; 103. a communication interface.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures.
It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
In the description of the embodiments of the present invention, "plurality" means at least 2.
Examples:
1-5, in a first aspect, an embodiment of the present invention provides a marine wind power identification method based on yolo algorithm, including the following steps:
s1, acquiring remote sensing images of offshore wind power plants in a plurality of areas, and manufacturing an offshore wind power sample set;
s2, constructing an initial offshore wind power identification model based on a YOLOv5 algorithm, a CA attention mechanism and an RFB multi-branch convolution module;
s3, training and verifying the initial offshore wind power identification model by utilizing the offshore wind power sample set to obtain a target offshore wind power identification model;
and S4, acquiring a real-time offshore wind power image, and identifying an offshore wind power target by using a target offshore wind power identification model.
The invention creates a deep learning training data set suitable for Chinese offshore wind power target recognition, optimizes and improves based on a YOLOv5 algorithm in a YOLO algorithm, introduces a CA attention mechanism and an RFB multi-branch convolution module, builds an effective offshore wind power recognition model (YOLOv 5s-CR model), adds a coordinate attention CA module to improve the positioning capability of the model, and combines a context enhancement module RFB to increase receptive fields to reduce the false detection rate of the model under a large-scale complex sea surface background, thereby realizing quick detection and accurate positioning of Chinese offshore wind power. The offshore wind power is accurately and stably detected based on the constructed offshore wind power identification model, and the detection efficiency and the detection accuracy are effectively improved. The marine wind power identification model constructed by the method has good generalization and geographic generalization performances.
Based on the first aspect, the method for manufacturing the offshore wind power sample set further comprises the following steps:
preprocessing remote sensing images of the offshore wind farm;
performing mean value calculation on the preprocessed remote sensing image of the offshore wind farm, and optimizing the remote sensing image of the offshore wind farm;
sample marking is carried out on the optimized remote sensing image of the offshore wind farm so as to obtain an offshore wind power sample test set and an offshore wind power sample verification set;
manufacturing sample slices of an offshore wind power sample test set and an offshore wind power sample verification set through an ArcGIS Pro image processing tool, and writing position information and category information of all samples into corresponding xml files;
and selecting part of sample slice data as a training set and a testing set to form a VOC data set.
Based on the first aspect, the method for preprocessing the remote sensing image of the offshore wind farm further comprises the following steps:
updating the orbit state vector of the remote sensing image of the offshore wind farm, and updating orbit metadata by using the corrected orbit file aiming at the ground distance multi-view image GRD in the remote sensing image of the offshore wind farm;
the method comprises the steps of inhibiting low-intensity noise and invalid data of the edge of a GRD scene through a low-value pixel mask, and eliminating additional noise in sub-strips of a remote sensing image of an offshore wind farm;
and calculating the back scattering intensity by adopting sensor calibration parameters in the metadata to perform radiation calibration, and performing orthographic correction on the remote sensing image of the offshore wind farm by combining a digital elevation model.
In some embodiments of the present invention, offshore wind power is represented as high brightness spindle-like or oblong bright spots on high resolution SAR remote sensing images due to its unique three-bladed rigid polyhedral morphology. Contrast graphically with the weakly back-scattered sea surface. Thus, the C-band, spatially resolved 10 meter Sentinel-1 sarrd product was used herein to select an Interference Wide (IW) band mode and a vertical-vertical (VV) polarization mode. The product may be obtained through a GoogleEarthEngine (GEE) platform. The Sentinel-1 satellite consists of two polar orbiting satellites A and B. The dual-star cooperative work mode shortens the revisitation period of Sentinel-1 from 12 days to 6 days, and can provide all-weather observation. The actual manufacturing flow of the sample is as follows. Step1, sentinel-1 image preprocessing. Updating a track state vector of the SAR image, and updating track metadata by using a corrected track file aiming at a ground distance multi-view image (GRD); suppressing low-intensity noise and invalid data of the GRD scene edge by masking the low-value pixel; eliminating additional noise in the sub-bands to reduce discontinuities between different sub-bands in the multi-band acquisition mode; and calculating the back scattering intensity by adopting sensor calibration parameters in the metadata to perform radiation calibration, and performing orthographic correction on the image by combining Digital Elevation Model (DEM) data to eliminate the influence of terrain. Step2, average calculation and downloading are carried out on SAR images in the study area of years 2015-2022 12 each year. The aspect ratio of the image slices remains uniform and is not less than 2048 pixels. Step3, manually marking the Chinese offshore wind power in the sample area (fig. 3 (b) - (d)), and obtaining 2143 offshore wind power samples (test set) and 744 offshore wind power samples (verification set) respectively. Step4, making a 256×256-pixel sample slice by an ArcGISPro image processing tool, and writing the position information and the category information of the sample into a corresponding xml file. Step5, taking the output 2991 pieces of sample slice data in 2022 as a training set and 771 pieces of slice data in 2021 as a test set, and forming a data set in the VOC format.
Based on the first aspect, the marine wind power identification method based on the yolo algorithm further comprises the following steps:
and performing image enhancement processing on the images in the data set.
The generalization capability of the enhancement model is enhanced, the image data enhancement is carried out on the data set, and the random brightness enhancement, the random contrast enhancement, the sharpening and the Gamma enhancement are carried out on 2991 original Chinese offshore wind powers of the training set.
Based on the first aspect, the method for training and verifying the initial offshore wind power identification model by using the offshore wind power sample set further comprises the following steps:
and training and verifying the initial offshore wind power identification model by the training set and the testing set respectively.
Based on the first aspect, the method for constructing the initial offshore wind power identification model based on the YOLOv5 algorithm, the CA attention mechanism and the RFB multi-branch convolution module further comprises the following steps:
introducing a CA attention mechanism into a backlight module of the YOLOv5 algorithm to construct a CS-CA module;
introducing an RFB multi-branch convolution module into a main network of the YOLOv5 algorithm;
and constructing an initial offshore wind power identification model based on the YOLOv5 algorithm, the CS-CA module and the RFB multi-branch convolution module.
The method comprises the steps of (1) in the Input module, adopting translation, sharpening, contrast enhancement and the like to improve model generalization capability and robustness by a YOLOv5s-CR model, and (2) in the step (2), the background module comprises Conv, coT module and SPFF module and is responsible for extracting image features, and in the background module, a CA attention mechanism is added in a C3 module to strengthen a visual Backbone, improve positioning capability and realize better performance, and in the step (3), the Neck network uses a path aggregation network (PANet enhanced feature information to improve detection performance of small targets, in addition, the RFB module is introduced to expand a sense field, capture feature information and control the ratio between the sense field size and the eccentricity to generate a space array of sense, and in the step (4), the sense field module is mainly a prediction field with border frame position, frame confidence, class probability, and the prediction field model is improved in the accuracy, and the prediction field model is shown in the scheme 2.
One of the main challenges facing the field of remote sensing target detection is to deal with the problem of target identification and small target false alarm in a complex background. The invention introduces a coordinate attention (CoordinateAttention, CA) module in the C3 module to significantly improve the capability of identifying the extraction and positioning of the offshore wind power features. The design of the module skillfully merges the position information of the Chinese offshore wind power and the high sensitivity of the relation between channels, the position information is embedded into the attention of the channels, the wider context information can be obtained, and the structure of the CA module is shown in the figure 2 (b). And carrying out average pooling operation on each channel of the input feature map X along the horizontal and vertical directions by using pooling kernels with the sizes of (H, 1) and (1, W), and encoding space coordinate information. This operation aggregates features in two spatial directions, respectively, yielding a direction-aware feature map and, respectively, models long-term dependencies in the spatial direction and preserves accurate position information in the other spatial direction. . Intermediate feature map F is formed by F1 operation (1 x1 convolution kernel dimension reduction) and nonlinear activation, and then is decomposed into a high attention tensor and a wide attention tensor. And performing dimension lifting operation through 21 multiplied by 1 convolutions, reducing complexity and calculation cost of a model by combining a sigmoid function, obtaining a height and width attention vector sum, and finally multiplying the input feature map X with the height and width attention weight sum respectively to generate a final attention feature map Y. The C3-CA module integrates the channel and the position information, and uses a attention mechanism to improve the detection performance of the offshore wind power in China and reduce the false alarm rate. An effective solution is provided for solving the problem of remote sensing target detection in the complex SAR background, and the calculation cost is hardly increased.
In order to improve the speed and accuracy in the Chinese offshore wind power target detection task, the invention introduces a RFB (ReceptiveFieldBlock) module. The multi-branch convolution layer is a core component of the RFB module, skillfully adopts a bottleneck structure, reduces the dimension of channel characteristics through the convolution layer with the 1x1 step of 2, and then introduces the convolution layer of nxn. To effectively control the number of parameters and increase the depth of the nonlinear layer, the module replaces the 5x5 convolution with 2 3x3 convolutions while introducing a 1x1 convolution layer without an activation function in the direct-connect layer. In addition, the RFB module controls the eccentricity of each branch through the dilation convolution technique, simulates the ratio between the size of the receptive field and the eccentricity, and finally, serially connects and carries out 1x1 convolution to generate a spatial array of receptive fields. A higher resolution representation of the feature is obtained while keeping the number of parameters small. The multiple branches fused with different convolution kernel sizes and expansion factors further improve the feature extraction capability, and provide powerful support for improving the performance of the module. The RFB module is integrated into a main network of the lightweight YOLOv5s, so that the quality of characteristic representation can be effectively improved, and the relationship between the receptive field size and the eccentricity can be accurately measured, so that the method is more differentiated and robust. Compared with the traditional deep backbone network, the RFB module introduces a more flexible feature extraction mechanism, and realizes the acceleration of tasks while ensuring the accuracy.
In other embodiments of the present invention, in order to further verify the performance of the proposed YOLOv5s-CR model, SAR image datasets of chinese offshore wind power from 2015 to 2022 were tested year by year in a time series manner. The test data set selected includes different sea areas, various underlying conditions, different fan types, water depths and offshore distances, and SAR images of complex land and sea backgrounds that have not been segmented by sea and land. We have paid particular attention to typical offshore wind farms in Jiangsu tidal flat, and have shown the identification visualization results of the offshore wind power in each year from 2015 to 2022 on the basis of time series, specifically in (a) - (h) in fig. 3, the red boxes represent the offshore wind power, and (a) - (h) represent the detection samples in 2015-2022, respectively. The result shows that the model can almost identify the Chinese offshore wind power targets in each year without omission, and the situation that the adjacent offshore wind power targets are misjudged to be the same target or the single Chinese offshore wind power targets are misjudged to be two targets never occurs. In addition, under the complex background that sea and land segmentation is not passed, the model can still accurately identify the Chinese offshore wind power. The experimental result comprehensively verifies the generalization capability of the SAR image of the model at different times, thereby further proving that the SAR image has the characteristic of high credibility. The YOLOv5s-CR model continues to perform excellently in time series experiments, fully demonstrating its excellent performance in terms of generalization in time and highlighting the potential for robustness and sustainable optimization.
In order to deeply study the geographic generalization of the YOLOv5s-CR model, we have expanded the application range of the model for typical offshore wind farm identification detection in other countries around the world, including multiple countries in the united kingdom, germany, denmark, france, belgium, etc. By means of model identification on SAR image data of the offshore wind farm in different geographic environments, stability and wide adaptability of the model in the global scope are verified.
The method comprises the steps of selecting representative SAR image samples from offshore wind farm data of each country to manufacture test sets with different scales. None of these images passed through the sea Liu Fenge in order to more fully investigate the applicability of the model. We applied YOLOv5s-CR model to these selected image data to perform offshore wind targets in an automated fashion, with partial detection results shown in fig. 4 (a) - (n). Through analysis of the identification result, the YOLOv5s-CR model is observed to accurately identify most of offshore wind power. However, there are some cases of missed detection (fig. 4 (d)), and the complex images of land and sea are more prone to false detection, and the buildings on land (fig. 4 (l)) and the dams in sea (fig. 4 (m)) interfere with model detection. This phenomenon may be related to noise interference present in the images and the complex interaction of the wind farm with the surrounding environment. The model test and analysis are comprehensively carried out on the offshore wind farm data sets of other countries around the world, and the YOLOv5s-CR model is considered to have higher identification precision and performance stability under different geographic environments, and has higher geographic generalization and wide applicability.
As shown in fig. 6, in a second aspect, an embodiment of the present invention provides an offshore wind power identification system based on yolo algorithm, which includes a sample set making module 100, an initial model building module 200, a model training and verifying module 300, and an offshore wind power identification module 400, wherein:
the sample set manufacturing module 100 is used for acquiring remote sensing images of the offshore wind farm in a plurality of areas and manufacturing an offshore wind power sample set;
the initial model building module 200 is used for building an initial offshore wind power identification model based on a YOLOv5 algorithm, a CA attention mechanism and an RFB multi-branch convolution module;
the model training and verifying module 300 is configured to train and verify the initial offshore wind power identification model by using the offshore wind power sample set, so as to obtain a target offshore wind power identification model;
the offshore wind power identification module 400 is configured to acquire a real-time offshore wind power image, and identify an offshore wind power target by using a target offshore wind power identification model.
The system is used for accurately and stably detecting the offshore wind power through a plurality of modules such as the sample set manufacturing module 100, the initial model construction module 200, the model training and verifying module 300, the offshore wind power identification module 400 and the like, and the constructed offshore wind power identification model effectively improves the detection efficiency and the detection accuracy. The invention creates a deep learning training data set suitable for Chinese offshore wind power target recognition, optimizes and improves based on a YOLOv5 algorithm in a YOLO algorithm, introduces a CA attention mechanism and an RFB multi-branch convolution module, builds an effective offshore wind power recognition model (YOLOv 5s-CR model), adds a coordinate attention CA module to improve the positioning capability of the model, and combines a context enhancement module RFB to increase receptive fields to reduce the false detection rate of the model under a large-scale complex sea surface background, thereby realizing quick detection and accurate positioning of Chinese offshore wind power. The offshore wind power is accurately and stably detected based on the constructed offshore wind power identification model, and the detection efficiency and the detection accuracy are effectively improved. The marine wind power identification model constructed by the method has good generalization and geographic generalization performances.
As shown in fig. 7, in a third aspect, an embodiment of the present application provides an electronic device, which includes a memory 101 for storing one or more programs; a processor 102. The method of any of the first aspects described above is implemented when one or more programs are executed by the processor 102.
And a communication interface 103, where the memory 101, the processor 102 and the communication interface 103 are electrically connected directly or indirectly to each other to realize data transmission or interaction. For example, the components may be electrically connected to each other via one or more communication buses or signal lines. The memory 101 may be used to store software programs and modules that are stored within the memory 101 for execution by the processor 102 to perform various functional applications and data processing. The communication interface 103 may be used for communication of signaling or data with other node devices.
The Memory 101 may be, but is not limited to, a random access Memory (Random Access Memory, RAM), a Read Only Memory (ROM), a programmable Read Only Memory (Programmable Read-Only Memory, PROM), an erasable Read Only Memory (Erasable Programmable Read-Only Memory, EPROM), an electrically erasable Read Only Memory (Electric Erasable Programmable Read-Only Memory, EEPROM), etc.
The processor 102 may be an integrated circuit chip with signal processing capabilities. The processor 102 may be a general purpose processor including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), etc.; but also digital signal processors (Digital Signal Processing, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), field programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
In the embodiments provided in the present application, it should be understood that the disclosed method and system may be implemented in other manners. The above-described method and system embodiments are merely illustrative, for example, flow charts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of methods and systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, the functional modules in the embodiments of the present application may be integrated together to form a single part, or each module may exist alone, or two or more modules may be integrated to form a single part.
In a fourth aspect, embodiments of the present application provide a computer-readable storage medium having stored thereon a computer program which, when executed by the processor 102, implements a method as in any of the first aspects described above. The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
It will be evident to those skilled in the art that the present application is not limited to the details of the foregoing illustrative embodiments, and that the present application may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Claims (9)
1. The marine wind power identification method based on the yolo algorithm is characterized by comprising the following steps of:
collecting remote sensing images of offshore wind farms in a plurality of areas, and manufacturing an offshore wind power sample set;
constructing an initial offshore wind power identification model based on a YOLOv5 algorithm, a CA attention mechanism and an RFB multi-branch convolution module;
training and verifying the initial offshore wind power identification model by utilizing the offshore wind power sample set to obtain a target offshore wind power identification model;
and acquiring a real-time offshore wind power image, and identifying an offshore wind power target by using a target offshore wind power identification model.
2. The method for identifying the offshore wind power based on the yolo algorithm according to claim 1, wherein the method for manufacturing the offshore wind power sample set comprises the following steps:
preprocessing remote sensing images of the offshore wind farm;
performing mean value calculation on the preprocessed remote sensing image of the offshore wind farm, and optimizing the remote sensing image of the offshore wind farm;
sample marking is carried out on the optimized remote sensing image of the offshore wind farm so as to obtain an offshore wind power sample test set and an offshore wind power sample verification set;
manufacturing sample slices of an offshore wind power sample test set and an offshore wind power sample verification set through an ArcGIS Pro image processing tool, and writing position information and category information of all samples into corresponding xml files;
and selecting part of sample slice data as a training set and a testing set to form a VOC data set.
3. The marine wind power identification method based on yolo algorithm according to claim 2, wherein the method for preprocessing the remote sensing image of the marine wind power plant comprises the following steps:
updating the orbit state vector of the remote sensing image of the offshore wind farm, and updating orbit metadata by using the corrected orbit file aiming at the ground distance multi-view image GRD in the remote sensing image of the offshore wind farm;
the method comprises the steps of inhibiting low-intensity noise and invalid data of the edge of a GRD scene through a low-value pixel mask, and eliminating additional noise in sub-strips of a remote sensing image of an offshore wind farm;
and calculating the back scattering intensity by adopting sensor calibration parameters in the metadata to perform radiation calibration, and performing orthographic correction on the remote sensing image of the offshore wind farm by combining a digital elevation model.
4. The marine wind power identification method based on yolo algorithm according to claim 2, further comprising the steps of:
and performing image enhancement processing on the images in the data set.
5. The method for identifying the offshore wind power based on the yolo algorithm according to claim 2, wherein the method for training and verifying the initial offshore wind power identification model by utilizing the offshore wind power sample set comprises the following steps:
and training and verifying the initial offshore wind power identification model by the training set and the testing set respectively.
6. The marine wind power identification method based on the yolo algorithm according to claim 1, wherein the method for constructing an initial marine wind power identification model based on the yolo v5 algorithm, the CA attention mechanism and the RFB multi-branch convolution module comprises the following steps:
introducing a CA attention mechanism into a backlight module of the YOLOv5 algorithm to construct a CS-CA module;
introducing an RFB multi-branch convolution module into a main network of the YOLOv5 algorithm;
and constructing an initial offshore wind power identification model based on the YOLOv5 algorithm, the CS-CA module and the RFB multi-branch convolution module.
7. The marine wind power identification system based on the yolo algorithm is characterized by comprising a sample set making module, an initial model building module, a model training and verifying module and a marine wind power identification module, wherein:
the sample set manufacturing module is used for acquiring remote sensing images of the offshore wind farm in a plurality of areas and manufacturing an offshore wind power sample set;
the initial model building module is used for building an initial offshore wind power identification model based on a YOLOv5 algorithm, a CA attention mechanism and an RFB multi-branch convolution module;
the model training and verifying module is used for training and verifying the initial offshore wind power identification model by utilizing the offshore wind power sample set so as to obtain a target offshore wind power identification model;
and the offshore wind power identification module is used for acquiring real-time offshore wind power images and identifying an offshore wind power target by utilizing the target offshore wind power identification model.
8. An electronic device, comprising:
a memory for storing one or more programs;
a processor;
the method of any of claims 1-6 is implemented when the one or more programs are executed by the processor.
9. A computer readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the method according to any of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311697617.9A CN117671504A (en) | 2023-12-11 | 2023-12-11 | Marine wind power identification method and system based on yolo algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311697617.9A CN117671504A (en) | 2023-12-11 | 2023-12-11 | Marine wind power identification method and system based on yolo algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117671504A true CN117671504A (en) | 2024-03-08 |
Family
ID=90069586
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311697617.9A Pending CN117671504A (en) | 2023-12-11 | 2023-12-11 | Marine wind power identification method and system based on yolo algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117671504A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118279751A (en) * | 2024-05-31 | 2024-07-02 | 华东师范大学 | Offshore wind turbine installed capacity calculation method and system based on Sentinel-2 image and Unet model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115880642A (en) * | 2022-12-29 | 2023-03-31 | 中国华能集团清洁能源技术研究院有限公司 | Marine target proximity degree early warning method and system based on machine vision |
| CN117152625A (en) * | 2023-08-07 | 2023-12-01 | 西安电子科技大学 | Remote sensing small target identification method, system, equipment and medium based on CoordConv and Yolov5 |
-
2023
- 2023-12-11 CN CN202311697617.9A patent/CN117671504A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115880642A (en) * | 2022-12-29 | 2023-03-31 | 中国华能集团清洁能源技术研究院有限公司 | Marine target proximity degree early warning method and system based on machine vision |
| CN117152625A (en) * | 2023-08-07 | 2023-12-01 | 西安电子科技大学 | Remote sensing small target identification method, system, equipment and medium based on CoordConv and Yolov5 |
Non-Patent Citations (4)
| Title |
|---|
| WENXIAO ZHAO 等: ""CRAS-YOLO：A Novel multi-category vessel detection and classification model based on YOLOv5s algorithm"", 《IEEE》, 19 February 2023 (2023-02-19) * |
| XINGSHUAI YANG 等: ""Detection of river floating garbage based on improved YOLOV5"", 《MATHEMATICS》, 20 November 2022 (2022-11-20) * |
| YINGCHAO ZHANG 等: ""converge of coordinate attention boosted yolov5 model and quantum dot labeled fluorescent biosensing for rapid detection of the poultry disease"", 《COMPUTER ANDELECTRONICS IN AGRICULTURE》, 15 February 2023 (2023-02-15) * |
| 朱昭云 等: ""基于YOLOv5s的海上风电设施检测与预警评估"", 《计算机技术与发展》, 1 April 2023 (2023-04-01) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118279751A (en) * | 2024-05-31 | 2024-07-02 | 华东师范大学 | Offshore wind turbine installed capacity calculation method and system based on Sentinel-2 image and Unet model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110705457B (en) | Remote sensing image building change detection method | |
| CN112183432B (en) | Building area extraction method and system based on medium-resolution SAR image | |
| Cooke et al. | Estimating sea ice concentration from SAR: Training convolutional neural networks with passive microwave data | |
| CN104613944A (en) | Distributed water depth prediction method based on GWR (geographically weighted regression) and BP (back propagation) neural network | |
| Castelle et al. | Primary drivers of multidecadal spatial and temporal patterns of shoreline change derived from optical satellite imagery | |
| CN103226826B (en) | Based on the method for detecting change of remote sensing image of local entropy visual attention model | |
| CN115546656B (en) | Remote sensing image cultivation region extraction method based on deep learning | |
| CN117671504A (en) | Marine wind power identification method and system based on yolo algorithm | |
| CN111598098A (en) | Water gauge water line detection and effectiveness identification method based on full convolution neural network | |
| Bai et al. | An intelligent water level monitoring method based on SSD algorithm | |
| Dang et al. | A convolutional neural network for coastal classification based on ALOS and NOAA satellite data | |
| CN112180369B (en) | Depth learning-based sea surface wind speed inversion method for one-dimensional synthetic aperture radiometer | |
| CN109583412A (en) | A kind of training method and its ship detecting method carrying out ship detecting using convolutional neural networks | |
| Chen et al. | Rain-contaminated region segmentation of X-band marine radar images with an ensemble of SegNets | |
| CN113628180A (en) | Semantic segmentation network-based remote sensing building detection method and system | |
| CN114119621A (en) | SAR remote sensing image water area segmentation method based on depth coding and decoding fusion network | |
| CN107463944B (en) | A kind of road information extracting method using multidate High Resolution SAR Images | |
| CN117437523B (en) | Weak trace detection method combining SAR CCD and global information capture | |
| CN114387446A (en) | Automatic water body extraction method for high-resolution remote sensing image | |
| CN103886289B (en) | Direction self-adaptive method and system for identifying on-water bridge targets | |
| CN115877345B (en) | Method and device for supplementing missing measurement data of wind profile radar | |
| CN112200845A (en) | Image registration method and device | |
| CN111951204A (en) | Sea surface wind speed inversion method for Tiangong No. two detection data based on deep learning | |
| CN116883594A (en) | Ocean three-dimensional temperature remote sensing super-resolution reconstruction method integrating satellite and buoy observation | |
| CN113096129B (en) | Method and device for detecting cloud cover in hyperspectral satellite image |
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 |