CN108256634A - A kind of ship target detection method based on lightweight deep neural network - Google Patents
A kind of ship target detection method based on lightweight deep neural network Download PDFInfo
- Publication number
- CN108256634A CN108256634A CN201810128120.8A CN201810128120A CN108256634A CN 108256634 A CN108256634 A CN 108256634A CN 201810128120 A CN201810128120 A CN 201810128120A CN 108256634 A CN108256634 A CN 108256634A
- Authority
- CN
- China
- Prior art keywords
- ship target
- channel
- model
- neural network
- deep neural
- 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
Classifications
-
- 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
Abstract
The invention discloses a kind of ship target detection methods based on lightweight deep neural network, include the following steps:Step S1:It designs and trains to obtain the lightweight deep neural network model of ship target detection;Step S2:Lightweight deep neural network model trained in step S1 is migrated into embedded device, acquires ship video in real time by camera, is detected by trained model to complete ship target.Technical solution using the present invention using completely new channel beta pruning Web compression technology, realizes the Ship Target Detection technology based on lightweight deep neural network, has many advantages, such as that device dependence is low, accuracy is high and real-time.
Description
Technical field
The present invention relates to the ship target detection methods on sea more particularly to one kind to be based on lightweight deep neural network
Ship target detection method.
Background technology
It is well known that China's marine economic industry activity is become increasingly prosperous, Resources of International Deep Sea-bed contention is increasingly fierce, especially
Naval of China builds gradually to be changed to ocean naval, and quick, accurate, light weight, system ship target detecting system is more and more aobvious
It obtains important.
Ship target detection is always aspect important during land and sea border defense is built, it is ensured that accurately and fast identifying
Sea ship target, while be alternatively commander's decision and support is provided.Accurately and fast detect the ship target identification on sea
It plays an important role in the action for consolidating land and sea border defense.
It is automatic that ship target detection method more common at present includes classical statistical pattern recognition method, Knowledge based engineering
Target identification method, the automatic target recognition method based on model, the automatic target recognition method based on artificial neural network, base
In the automatic target recognition method of artificial intelligence and deep learning.Since the ship target of statistical pattern recognition method detects skill
Art, to complicated pattern extraction feature difficulty, accuracy in practice is all relatively low;The ship mesh of knowledge based, model
What the method rule of mark detection technique was often rule of thumb got, accurate rational model is often difficult to set up, so as to also limit
Its application is made;Although the ship target identification technology based on Artificial Neural Network has certain promotion in accuracy rate,
It is that network depth is shallower, can not fully extracts very much feature;It is identified based on the ship target of artificial intelligence and deep learning
Method can reach good accuracy of detection using depth convolutional neural networks, but still be weak in terms of real-time, and
It is more demanding to hardware device resources, it is normally operated in server, can not be applied in embedded device.
Invention content
For technical problem of the existing technology, the present invention proposes a kind of ship based on lightweight deep neural network
Object detection method.The advantage of the invention is to combine NP-Hard Resolving probiems thinkings, designs distinctive feature channel beta pruning side
Method, lightweight deep neural network enable to apply in the embedded device of lower-performance, and there is relatively low equipment to rely on
Property, while while keeping compared with high-accuracy, moreover it is possible to network calculations amount is reduced, so as to improve real-time.
To solve its technical problem, the technical solution adopted in the present invention is as follows:
A kind of ship target detection method based on lightweight deep neural network, which is characterized in that include the following steps:
Step S1:It designs and trains to obtain the lightweight deep neural network model of ship target detection;
Step S2:Lightweight deep neural network model trained in step S1 is migrated into embedded device, is passed through
Camera acquires ship video in real time, is detected by trained model to complete ship target;
Wherein, step S1 further comprises the steps:
Step S11:Make two datasets:First data set is trained for sorter network, according to the form of Imagenet
The specified data set of generation;Second data set is labeled the ship target image of all acquisitions, obtains for detecting network training
Co-ordinate position information and classification information of the corresponding ship target in figure are obtained, and generates the XML file of specified format;
Step S12:According to the first data set, ship target base categories are carried out using depth convolutional neural networks model
Model training obtains original classification network model;
Step S13:To the original classification network model that step S12 is obtained, using following steps channel pruning method into
Row processing:
Step S131:Design acceleration assessment algorithm
Wherein, X is the input feature vector of N-dimensional, XiIt is i-th of channel of X, YiIt is i-th of channel of the corresponding convolution kernel groups of X,
θ is the enabled vector of channel selecting, the length of c;θ i=0, then represent XiChannel is by beta pruning;θ i=1, then represent XiIt is logical
Road is not by beta pruning;λ is penalty coefficient;
Step S132:By the algorithm evaluation individual layer compression performance in step S131, beta pruning optimization problem is regarded as NP-
Hard problems carry out parameter optimization, realize the channel beta pruning of single layer network;
Step S133:It is cut layer by layer into row of channels in order with the method for the channel beta pruning of single layer network in step S132
Branch, completes the channel beta pruning of global network, obtains lightweight ship target Fast Classification model;
Step S14:The lightweight ship target Fast Classification model that will be obtained in step S13, is transplanted to Faster R-
It in the characteristic extracting module of CNN (the target detection model of a kind of mainstream), is then trained, obtains detecting for ship target
Lightweight deep neural network model.
Compared with prior art, the beneficial effects of the invention are as follows:To the ocean for having ship target of camera actual acquisition
Image carries out target detection, in the premise for not reducing original deep neural network precision by the deep neural network of lightweight
Under, ship target detection identification is completed more quickly, and network rapidity greatly improves.Using distinctive channel technology of prunning branches, have
Effect reduces model parameter, model size is reduced, so as to reduce calculation amount.With the Web compressions method phase such as existing weights beta pruning
Than the channel pruning method designed by the present invention has abandoned the cut operator of that original violence, then utilizes mathematical method excellent
Change object function, using the strategy of structuring beta pruning, eliminate the reliance on specific sparse convolution and calculate library, while high pressure can obtained
Greatly reduce the calculating time of test phase while shrinkage.And it can make on the embedded platform of relatively low calculated performance
With greatly improving the popularity used.
Description of the drawings
Fig. 1 is the flow chart of the ship target detection method the present invention is based on lightweight deep neural network.
Specific embodiment
Below in conjunction with attached drawing, the present invention will be further described.
In view of the defects existing in the prior art, the present invention proposes a kind of target detection based on lightweight deep neural network
Method can preferably break away from dependence of the conventional depth learning method to high-capability computing device, can be used effectively in
In some embedded devices, the testing requirements compared with high target are realized with smaller network model.
In order to train lightweight deep neural network model, required image is first gathered:10 kinds of different types of ships
(aircraft carrier, submarine, medical care ship, fisheries administration ship, fishing boat, cruise, lifeboat, inflatable boat, disappears at escort vessel navigation image across the sea
Anti- ship) each classification each 1000, in total 10000, picture.Referring to Fig. 1, it show refreshing the present invention is based on lightweight depth
The flow chart of ship target detection method through network, specific experimental procedure are as follows:
Step S1:It is marked by dataset acquisition, original classification network training, and compresses to obtain light weight by channel beta pruning
Grade ship target base categories model, then the characteristic extracting module for replacing Faster R-CNN is transplanted, finally fine tuning training obtains
The lightweight deep neural network model of ship target detection;
Step S2:Lightweight deep neural network model trained in step S1 is migrated into embedded device, is passed through
Camera acquires ship video in real time, is detected by trained model to complete ship target;
Wherein, step S1 further comprises the steps:
Step S11:Make two datasets:First data set is trained for sorter network, according to the form of Imagenet
The specified data set of generation;Second data set is labeled the ship target image of all acquisitions, obtains for detecting network training
Co-ordinate position information and classification information of the corresponding ship target in figure are obtained, and generates the XML file of specified format;More than two
Class data set all presses 3 respectively:1 ratio cut partition is into training set and test set;
Step S12:According to the first data set, ship target base categories are carried out using depth convolutional neural networks model
Model training, basic convolutional neural networks are made of 13 convolutional layers, convolutional layer with multiple dimensioned convolution unit, Relu (assuming that
Function) unit, maximum pond layer and normalization layer.Contain 3 full articulamentums and a 10 types (ship mesh in sorter network
Mark categorized data set classification) softmax graders.In order to enable training is faster, quick GPU operations is used to perform convolution
Operation.In order to reduce the over-fitting of full articulamentum, the regular method of entitled dropout (neuron Random Activation) is employed,
Finally obtain original classification network model;
Step S13:To the original classification network model that step S12 is obtained, using the method for distinctive channel beta pruning, find out
And retaining the representative channel of each layer, and remaining channel is removed, specific channel beta pruning step is as follows:
Step S131:Design acceleration assessment algorithm
Wherein, X is the input feature vector of N-dimensional, XiIt is i-th of channel of X, YiIt is i-th of channel of the corresponding convolution kernel groups of X,
θ is the enabled vector of channel selecting, the length of c;θ i=0, then represent XiChannel is by beta pruning;θ i=1, then represent XiIt is logical
Road is not by beta pruning;λ is penalty coefficient, can increase the quantity of channel beta pruning by increasing λ, so as to improve compression ratio and acceleration
Than;
Step S132:The present invention can have an impact downstream layer, therefore to carry out after unessential channel is directly reduced
Network reconfiguration optimizes convolution kernel group parameter Y;By the algorithm evaluation individual layer compression performance in step S131, beta pruning is optimized
Problem is regarded as NP-Hard problems and carries out parameter optimization:First, fixed Y, re-optimization is to pick out most representational channel;
Then, fixed, re-optimization Y finally realizes the channel beta pruning of single layer network;
Step S133:It is cut layer by layer into row of channels in order with the method for the channel beta pruning of single layer network in step S132
Branch using the output of last layer as from current layer input characteristics figure, obtains the ship target Fast Classification model after channel beta pruning,
And it is trained fine tuning so that the front and rear network class Performance optimization of beta pruning compression is completed the channel beta pruning of global network, obtained
To lightweight ship target Fast Classification model, the progress fast target classification under the premise of classification accuracy is not reduced is realized;
Step S14:The lightweight ship target Fast Classification model that will be obtained in step S13, is transplanted to Faster R-
In the characteristic extracting module of CNN (the target detection model of a kind of mainstream), ship target inspection can be used for by being then trained to obtain
The lightweight deep neural network model of survey;In order to accelerate the speed of network training with improving neural network accuracy, in the training process,
It does not train from the beginning, but the parameter of sorter network model obtained using step S13 is used as feature extraction network
Partial initiation parameter, so as to faster and better obtain training result.
It is logical designed by the present invention compared with the Web compressions methods such as existing weights beta pruning using above-mentioned technical proposal
The strategy that road pruning method will use structuring beta pruning eliminates the reliance on specific sparse convolution and calculates library, while can obtain height
Greatly reduce the calculating time of test phase while compression ratio.The invention can be used for the sea between China and periphery sea neighbouring country
Foreign borderline region, the marine resources such as effective protection natural gas, oil, prevents precious resources to be ransacked of;Can also be used for protection territorial waters,
Military security helps command of armed force personnel to grasp marine ship information in time, so as to effectively reach accurate commander, accurate control,
Accurate collaboration, and make effective security decision.In contrast, the ship target inspection based on lightweight deep neural network
Survey technology by the Web compression technology of novel channel beta pruning, has that device dependence is low, accuracy is high and real-time etc.
Advantage.
Claims (1)
1. a kind of ship target detection method based on lightweight deep neural network, which is characterized in that include the following steps:
Step S1:It designs and trains to obtain the lightweight deep neural network model of ship target detection;
Step S2:Lightweight deep neural network model trained in step S1 is migrated into embedded device, passes through camera shooting
Head acquires ship video in real time, is detected by trained model to complete ship target;
Wherein, step S1 further comprises the steps:
Step S11:Make two datasets:First data set is trained for sorter network, is generated according to the form of Imagenet
Specified data set;Second data set is labeled the ship target image of all acquisitions, obtains phase for detecting network training
Co-ordinate position information and classification information of the ship target in figure are answered, and generates the XML file of specified format;
Step S12:According to the first data set, ship target base categories model is carried out using depth convolutional neural networks model
Training, obtains original classification network model;
Step S13:To the original classification network model that step S12 is obtained, at the channel pruning method using following steps
Reason:
Step S131:Design acceleration assessment algorithm
Wherein, X is the input feature vector of N-dimensional, XiIt is i-th of channel of X, YiIt is i-th of channel of the corresponding convolution kernel groups of X, θ is
The enabled vector of channel selecting, the length of c;θ i=0, then represent XiChannel is by beta pruning;θ i=1, then represent XiChannel is not
By beta pruning;λ is penalty coefficient;
Step S132:By the algorithm evaluation individual layer compression performance in step S131, beta pruning optimization problem is regarded as NP-Hard and is asked
Topic carries out parameter optimization, realizes the channel beta pruning of single layer network;
Step S133:With the method for the channel beta pruning of single layer network in step S132 in order layer by layer into row of channels beta pruning,
The channel beta pruning of global network is completed, obtains lightweight ship target Fast Classification model;
Step S14:The lightweight ship target Fast Classification model that will be obtained in step S13, is transplanted to Faster R-CNN (one
The target detection model of class mainstream) characteristic extracting module in, be then trained, obtain the light weight detected for ship target
Grade deep neural network model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810128120.8A CN108256634A (en) | 2018-02-08 | 2018-02-08 | A kind of ship target detection method based on lightweight deep neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810128120.8A CN108256634A (en) | 2018-02-08 | 2018-02-08 | A kind of ship target detection method based on lightweight deep neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108256634A true CN108256634A (en) | 2018-07-06 |
Family
ID=62743915
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810128120.8A Pending CN108256634A (en) | 2018-02-08 | 2018-02-08 | A kind of ship target detection method based on lightweight deep neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108256634A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109214353A (en) * | 2018-09-27 | 2019-01-15 | 云南大学 | A kind of facial image based on beta pruning model quickly detects training method and device |
CN109376580A (en) * | 2018-08-30 | 2019-02-22 | 杭州电子科技大学 | A kind of electric tower component identification method based on deep learning |
CN109509214A (en) * | 2018-10-15 | 2019-03-22 | 杭州电子科技大学 | A kind of ship target tracking based on deep learning |
CN109543766A (en) * | 2018-11-28 | 2019-03-29 | 钟祥博谦信息科技有限公司 | Image processing method and electronic equipment, storage medium |
CN109615858A (en) * | 2018-12-21 | 2019-04-12 | 深圳信路通智能技术有限公司 | A kind of intelligent parking behavior judgment method based on deep learning |
CN109858481A (en) * | 2019-01-09 | 2019-06-07 | 杭州电子科技大学 | A kind of Ship Target Detection method based on the detection of cascade position sensitivity |
CN109886114A (en) * | 2019-01-18 | 2019-06-14 | 杭州电子科技大学 | A kind of Ship Target Detection method based on cluster translation feature extraction strategy |
CN109919113A (en) * | 2019-03-12 | 2019-06-21 | 北京天合睿创科技有限公司 | Ship monitoring method and system and harbour operation prediction technique and system |
CN110119757A (en) * | 2019-03-28 | 2019-08-13 | 北京奇艺世纪科技有限公司 | Model training method, video category detection method, device, electronic equipment and computer-readable medium |
CN111062383A (en) * | 2019-11-04 | 2020-04-24 | 南通大学 | Image-based ship detection depth neural network algorithm |
CN111144208A (en) * | 2019-11-22 | 2020-05-12 | 北京航天控制仪器研究所 | Automatic detection and identification method for marine vessel target and target detector |
US10782691B2 (en) | 2018-08-10 | 2020-09-22 | Buffalo Automation Group Inc. | Deep learning and intelligent sensing system integration |
US10936907B2 (en) | 2018-08-10 | 2021-03-02 | Buffalo Automation Group Inc. | Training a deep learning system for maritime applications |
CN113269312A (en) * | 2021-06-03 | 2021-08-17 | 华南理工大学 | Model compression method and system combining quantization and pruning search |
CN113763484A (en) * | 2021-09-17 | 2021-12-07 | 交通运输部水运科学研究所 | Ship target positioning and speed estimation method based on video image analysis technology |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5636326A (en) * | 1992-09-04 | 1997-06-03 | Ricoh Corporation | Method for operating an optimal weight pruning apparatus for designing artificial neural networks |
WO2003079286A1 (en) * | 2002-03-15 | 2003-09-25 | Pacific Edge Biotechnology Limited | Medical applications of adaptive learning systems using gene expression data |
CN105512680A (en) * | 2015-12-02 | 2016-04-20 | 北京航空航天大学 | Multi-view SAR image target recognition method based on depth neural network |
CN107085733A (en) * | 2017-05-15 | 2017-08-22 | 山东工商学院 | Offshore infrared ship recognition methods based on CNN deep learnings |
CN107145903A (en) * | 2017-04-28 | 2017-09-08 | 武汉理工大学 | A kind of Ship Types recognition methods extracted based on convolutional neural networks picture feature |
CN107609601A (en) * | 2017-09-28 | 2018-01-19 | 北京计算机技术及应用研究所 | A kind of ship seakeeping method based on multilayer convolutional neural networks |
-
2018
- 2018-02-08 CN CN201810128120.8A patent/CN108256634A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5636326A (en) * | 1992-09-04 | 1997-06-03 | Ricoh Corporation | Method for operating an optimal weight pruning apparatus for designing artificial neural networks |
WO2003079286A1 (en) * | 2002-03-15 | 2003-09-25 | Pacific Edge Biotechnology Limited | Medical applications of adaptive learning systems using gene expression data |
CN105512680A (en) * | 2015-12-02 | 2016-04-20 | 北京航空航天大学 | Multi-view SAR image target recognition method based on depth neural network |
CN107145903A (en) * | 2017-04-28 | 2017-09-08 | 武汉理工大学 | A kind of Ship Types recognition methods extracted based on convolutional neural networks picture feature |
CN107085733A (en) * | 2017-05-15 | 2017-08-22 | 山东工商学院 | Offshore infrared ship recognition methods based on CNN deep learnings |
CN107609601A (en) * | 2017-09-28 | 2018-01-19 | 北京计算机技术及应用研究所 | A kind of ship seakeeping method based on multilayer convolutional neural networks |
Non-Patent Citations (3)
Title |
---|
QIAO SUN 等: "Method of Battlefield Target Information Sharing Based on Ontology", 《2017 4TH INTERNATIONAL CONFERENCE ON INFORMATION SCIENCE AND CONTROL ENGINEERING (ICISCE)》 * |
YIHUI HE 等: ""Channel Pruning for Accelerating Very Deep Neural Networks"", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
任陈俊: "基于机器视觉的场景目标检测与分类研究", 《中国优秀硕士论文全文数据库 信息科技辑》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10782691B2 (en) | 2018-08-10 | 2020-09-22 | Buffalo Automation Group Inc. | Deep learning and intelligent sensing system integration |
US10936907B2 (en) | 2018-08-10 | 2021-03-02 | Buffalo Automation Group Inc. | Training a deep learning system for maritime applications |
CN109376580A (en) * | 2018-08-30 | 2019-02-22 | 杭州电子科技大学 | A kind of electric tower component identification method based on deep learning |
CN109376580B (en) * | 2018-08-30 | 2022-05-20 | 杭州电子科技大学 | Electric power tower component identification method based on deep learning |
CN109214353A (en) * | 2018-09-27 | 2019-01-15 | 云南大学 | A kind of facial image based on beta pruning model quickly detects training method and device |
CN109214353B (en) * | 2018-09-27 | 2021-11-23 | 云南大学 | Training method and device for rapid detection of face image based on pruning model |
CN109509214A (en) * | 2018-10-15 | 2019-03-22 | 杭州电子科技大学 | A kind of ship target tracking based on deep learning |
CN109543766A (en) * | 2018-11-28 | 2019-03-29 | 钟祥博谦信息科技有限公司 | Image processing method and electronic equipment, storage medium |
CN109615858A (en) * | 2018-12-21 | 2019-04-12 | 深圳信路通智能技术有限公司 | A kind of intelligent parking behavior judgment method based on deep learning |
CN109858481A (en) * | 2019-01-09 | 2019-06-07 | 杭州电子科技大学 | A kind of Ship Target Detection method based on the detection of cascade position sensitivity |
CN109886114A (en) * | 2019-01-18 | 2019-06-14 | 杭州电子科技大学 | A kind of Ship Target Detection method based on cluster translation feature extraction strategy |
CN109919113A (en) * | 2019-03-12 | 2019-06-21 | 北京天合睿创科技有限公司 | Ship monitoring method and system and harbour operation prediction technique and system |
CN110119757A (en) * | 2019-03-28 | 2019-08-13 | 北京奇艺世纪科技有限公司 | Model training method, video category detection method, device, electronic equipment and computer-readable medium |
CN110119757B (en) * | 2019-03-28 | 2021-05-25 | 北京奇艺世纪科技有限公司 | Model training method, video category detection method, device, electronic equipment and computer readable medium |
CN111062383A (en) * | 2019-11-04 | 2020-04-24 | 南通大学 | Image-based ship detection depth neural network algorithm |
CN111144208A (en) * | 2019-11-22 | 2020-05-12 | 北京航天控制仪器研究所 | Automatic detection and identification method for marine vessel target and target detector |
CN113269312A (en) * | 2021-06-03 | 2021-08-17 | 华南理工大学 | Model compression method and system combining quantization and pruning search |
CN113763484A (en) * | 2021-09-17 | 2021-12-07 | 交通运输部水运科学研究所 | Ship target positioning and speed estimation method based on video image analysis technology |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108256634A (en) | A kind of ship target detection method based on lightweight deep neural network | |
CN107563303B (en) | Robust ship target detection method based on deep learning | |
CN110163108B (en) | Robust sonar target detection method based on dual-path feature fusion network | |
CN112766087A (en) | Optical remote sensing image ship detection method based on knowledge distillation | |
CN106887011B (en) | A kind of multi-template method for tracking target based on CNN and CF | |
CN109740460B (en) | Optical remote sensing image ship detection method based on depth residual error dense network | |
CN107609601A (en) | A kind of ship seakeeping method based on multilayer convolutional neural networks | |
CN108596030A (en) | Sonar target detection method based on Faster R-CNN | |
CN109886217A (en) | A method of it is high that wave being detected from Nearshore Wave video based on convolutional neural networks | |
CN108765458A (en) | High sea situation unmanned boat sea-surface target dimension self-adaption tracking based on correlation filtering | |
CN111091095B (en) | Method for detecting ship target in remote sensing image | |
CN108444447A (en) | A kind of fishing net in underwater obstacle avoidance system autonomous detection method in real time | |
CN109655815A (en) | Sonar target detection method based on SSD | |
CN110647802A (en) | Remote sensing image ship target detection method based on deep learning | |
CN109359661B (en) | Sentinel-1 radar image classification method based on convolutional neural network | |
CN107944386A (en) | Visual scene recognition methods based on convolutional neural networks | |
CN109886114A (en) | A kind of Ship Target Detection method based on cluster translation feature extraction strategy | |
CN107731011A (en) | A kind of harbour is moored a boat monitoring method, system and electronic equipment | |
CN110766721B (en) | Carrier landing cooperative target detection method based on airborne vision | |
Zhang et al. | A object detection and tracking method for security in intelligence of unmanned surface vehicles | |
CN114782817A (en) | Unmanned aerial vehicle water column target detection method based on improved YOLO X | |
CN115115863A (en) | Water surface multi-scale target detection method, device and system and storage medium | |
Chen et al. | Surface defect detection method based on improved attention mechanism and feature fusion model | |
CN116416136A (en) | Data amplification method for ship target detection of visible light remote sensing image and electronic equipment | |
Zhi-Feng et al. | Light-YOLOv3: fast method for detecting green mangoes in complex scenes using picking robots |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180706 |