CN109657541A - A kind of ship detecting method in unmanned plane image based on deep learning - Google Patents
A kind of ship detecting method in unmanned plane image based on deep learning Download PDFInfo
- Publication number
- CN109657541A CN109657541A CN201811330500.6A CN201811330500A CN109657541A CN 109657541 A CN109657541 A CN 109657541A CN 201811330500 A CN201811330500 A CN 201811330500A CN 109657541 A CN109657541 A CN 109657541A
- Authority
- CN
- China
- Prior art keywords
- ship
- deep learning
- unmanned plane
- network
- layer
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
Ship detecting method in the invention discloses a kind of unmanned plane image based on deep learning.Firstly, acquisition includes the unmanned plane image of military boats and ships and civilian ship and is labeled, ship datebase is obtained;Then, the ship datebase of acquisition is sent into deep learning network to be trained, until network convergence;Finally, detecting the ship target in unmanned plane image using trained deep learning network and weight file, and output test result.The present invention has preferable accuracy and robustness, efficiently solve the problems, such as the small target deteection in unmanned plane image, also solve the environmental disturbances that ship detecting is realized in traditional images Processing Algorithm simultaneously, illumination effect and the low problem of accurate rate, can be adapted for the ship detecting under different scenes.
Description
Technical field
The invention belongs to technical field of computer vision, in particular to a kind of unmanned plane figure based on deep learning
Ship detecting method as in.
Background technique
With the development of the social economy, communication is rapidly developing, ship increases year by year on quantity, tonnage and the speed of a ship or plane
Long, the frequency that ship collision, ship hit the generation of the accidents such as bridge is also higher and higher.The casualties as caused by navigation accident, wealth
It produces loss and environmental disruption is quite surprising;More importantly it is, with the fast development of industrial technology, civilian ship and military
The Division identification of ships is related to national security and people's safety problem, especially in the accurate positioning to marine external vessel.Ship
The detection of oceangoing ship can be widely used in military, civilian each field, in land resources investigation, seafari, military investigation, beat
Analysis and assessment etc. is hit to play an important role.Meanwhile it being increasingly enhanced with empty, balance table data retrieval capabilities, benefit
Ship target detection is carried out with unmanned air vehicle technique and identification is more and more paid attention to.The accurate positioning and identification of ship, can
To play positive effect in the safe navigation of ship and collision prevention, and can mention for water surface navigation channel for detecting monitoring ship
For relatively reliable data.In war under the conditions of Modern High-Tech, as early as possible to there is the military target seriously threatened to visit
It surveys and identification is also very important.
Spot ship detection algorithm mainly have ship detecting algorithm based on feature, view-based access control model ship detecting algorithm,
Ship detecting algorithm based on inter-frame difference and the ship detecting algorithm based on deep learning etc..Wherein view-based access control model and frame-to-frame differences
The ship detecting algorithm divided is mainly used for moving ship detection, and the ship detecting algorithm based on feature is more often available to still image
Detection, and the ship detecting algorithm based on deep learning is suitable for moving and static ship detecting.
Currently, usually requiring image preprocessing, image characteristics extraction, image point for the ship detecting algorithm based on feature
Class and etc..Wang Min et al. carries out automatic identification to large ship on high-resolution remote sensing image and extracts, article combination port
Mouth-shaped feature devises different algorithms and extracts, but this method is for large size according to the different operating statuses of ship
Ship and the apparent ship detecting effect of shape Small object ship that is good, but not being suitable in image.Peng Jingjing et al. is to boat
Clap the detecting and tracking algorithm that monitor video proposes a kind of multiple mobile object.Firstly, being filtered pre- place to the image sequence of extraction
Reason removal noise, recycles and detects target every frame frame difference method;The analysis of morphological operation and connected domain is carried out, then to overcome
Frame difference method is for the cavity occurred among longer target or slower moving object detection and the defect of inaccuracy;Finally using mentioning
The hull characteristics application Kalman filter taken tracks it.This method can effectively solve the occlusion issue of multiple target, have stronger
Robustness and application.But this method can not catch ship static or that navigation is very slow just for moving ship very well
It catches.Liu into et al. propose a kind of Ship Target based on Analysis On Multi-scale Features cluster and quickly position and recognition methods, use for reference the side Canny
The thought of edge detection and image pyramid and minimum spanning tree clustering algorithm devises and a set of is suitable for the multiple dimensioned characteristic of target simultaneously
Improve " feature cluster " target location algorithm of Canny edge detection adaptivity, join probability Tree Classifier and two-dimentional principal component
Analysis is calculated, and can be identified to multi-angle of view, multiple target type Ship Target, and assesses recognition result according to total probability formula, right
For big target in image, this method effect is pretty good, if naval vessel only accounts for several pixels of image, detection effect
It is not ideal.
Summary of the invention
To solve the problems, such as that spot ship detection algorithm depends on illumination, quality and the solution unmanned plane of image unduly
The fewer problem of pixel shared by ship in image, allows ship detecting to have better adaptability and applicability, and the present invention mentions
For a kind of ship detecting method in unmanned plane image based on deep learning.
In order to achieve the above technical purposes, the technical solution of the present invention is as follows:
A kind of ship detecting method in unmanned plane image based on deep learning, comprising the following steps:
(1) acquisition includes the unmanned plane image of military boats and ships and civilian ship and is labeled, and obtains ships data
Library;
(2) ship datebase that step (1) obtains is sent into deep learning network to be trained, until network convergence;
(3) it is detected in unmanned plane image using the trained deep learning network of step (2) and weight file
Ship target, and output test result.
Further, specific step is as follows for step (1):
(A1) the unmanned plane image comprising ship for training deep learning network is acquired;
(A2) acquired image is subjected to data prediction;
(A3) pretreated image is labeled with rectangle frame, obtains and is wrapped in the coordinate and rectangle frame of rectangle frame
Type containing ship, finally obtains ship datebase.
Further, in step (A2), the data prediction include lose image not comprising ship target and
Ship target shows less than half of image.
Further, it needs to mark two class data: military boats and ships and civilian ship in step (A3), uses floating-point first
Type data format records the location information of rectangle frame, is then converted into the data format that improved deep learning network needs.
Further, deep learning network described in step (2) is improved YOLT network, the knot of improved YOLT network
Structure is as shown in the table:
Training data or test data successively pass through 0,1,2 ... from the 0th layer of convolutional layer input network, 28 layers of processing, most
It is exported eventually from the 28th layer of convolutional layer;Wherein, at the 12nd layer Passthrough layers, the 6th layer of convolutional layer and 11th layer pond layer are connected
It picks up and, be then input to the 13rd layer of convolutional layer;At the 26th layer Passthrough layers, the 17th layer of convolutional layer and the 25th layer are rolled up
Lamination connects, and is then input to the 27th layer of convolutional layer.
Further, the improved YOLT network uses anchor boxes mechanism, using 7 anchors, and utilizes
K-means Mean Method obtains anchors size.
Further, specific step is as follows for step (2):
(B1) ship datebase is sent into deep learning network, chooses the good deep learning mould of ImageNet database training
Type is as basic network model, and the learning rate that training starts is 0.0001, and the parameter steps of percentage regulation learning network comes later
Change learning rate;
(B2) image in ship datebase is zoomed into fixed dimension at random, obtains training image;
(B3) training image is divided into n*n grid, 7 bounding boxes of each grid forecasting;
(B4) training image is sent into deep learning network, it is defeated after convolutional layer, pond layer, Passthrough layers
Convolution characteristic pattern out;
(B5) it is predicted with the convolution characteristic pattern of output using convolution, sliding window sampling is carried out on convolution characteristic pattern, according to
Step (B3), 7 different size of bounding boxes of each grid forecasting eventually predict n*n*7 bounding box, each bounding box
It returns position, confidence level and classification information and weight is adjusted by backpropagation then according to calculated loss function value;
(B6) terminate to train when reaching maximum number of iterations or when loss function value has reached convergence state, obtain final
Deep learning network and weight file for ship detecting.
Further, in step (B2), zoom to fixed dimension M*M, M ∈ [320,608], and step-length is 32;In step
Suddenly in (B3), n ∈ [10,19].
Further, in step (B5), the calculation method of loss function value is as follows:
One threshold value a is set, when averagely there is target weight value avg_obj to be less than threshold value a, lossnew=lossold/
(avg_obj+0.001), otherwise lossnew=lossold, wherein lossoldIt is the loss function of former YOLT network, lossnewIt is
The loss function of improved YOLT network.
Further, specific step is as follows for step (3):
(C1) Aerial Images to be detected are sent into trained deep learning network, will be taken photo by plane figure according to step (B2)
As zooming to fixed dimension, step step (B3), (B4) are repeated, final characteristic pattern is exported;
(C2) it is predicted with final characteristic pattern using convolution, sliding window sampling is carried out on convolution characteristic pattern, eventually in advance
N*n*7 bounding box is surveyed, each bounding box returns position, confidence level and classification information;
(C3) threshold value is set, the low bounding box of score is filtered out, non-maximum is carried out to the bounding box remained
Inhibition processing, obtains final testing result by classifier later, to realize the accurate positioning and identification of ship.
By adopting the above technical scheme bring the utility model has the advantages that
The present invention reduces complicated image preprocessing processes, reduce the requirement of the quality of image, and algorithm is simple, detection
High-efficient, detection accuracy is high, practical, has well adapting to property and reliability.The present invention efficiently solves illumination change
Change, extraction feature is not abundant, target is smaller and the interference of noise is influenced to ship detecting bring.
Detailed description of the invention
Fig. 1 is overview flow chart of the present invention;
Fig. 2 is collection process flow chart in the present invention;
Fig. 3 is training process flow chart in the present invention;
Fig. 4 is detection process flow chart in the present invention;
Fig. 5 is the unmanned plane figure with civilian ship target inputted in embodiment;
Fig. 6 is the unmanned plane figure with military boats and ships target inputted in embodiment;
Fig. 7 is the result figure of civilian ship detection in embodiment;
Fig. 8 is the result figure that military boats and ships detect in embodiment.
Specific embodiment
Below with reference to attached drawing, technical solution of the present invention is described in detail.
As shown in Figure 1, the ship detecting method in the unmanned plane image proposed by the present invention based on deep learning, according to
Secondary includes collection process, training process and detection process.
One, collection process: unmanned plane image of the acquisition comprising military boats and ships and civilian ship is simultaneously labeled, and is obtained
Ship datebase.The specific steps of collection process are as shown in Figure 2:
(A1) the unmanned plane image comprising ship for training deep learning network is acquired.
(A2) acquired image is subjected to data prediction, the image and ship of ship target is not included including losing
Target shows less than half of image.
(A3) pretreated image is labeled with rectangle frame, obtains and is wrapped in the coordinate and rectangle frame of rectangle frame
Type containing ship, finally obtains ship datebase.
To the mask method of image are as follows: two class data of label, military boats and ships and civilian ship are needed, for indicia framing position
Accuracy, the present embodiment using real-coded GA format recording mark frame location information (the indicia framing upper left corner and the lower right corner
Coordinate), it is then converted into the data format that improved deep learning network needs.The data file lattice that deep learning network needs
Formula is as follows:
<object-class><x><y><width><height>
Wherein, object-class is the index of class, and subsequent 4 values are relative to the ratio of whole picture.X is
The x coordinate at the center ROI, y are the y-coordinates at the center ROI, and width is the width of ROI, and height is the height of ROI.
Two, training process: the ship datebase of acquisition is sent into deep learning network and is trained, until network convergence.
Improved YOLT network is used in the present invention, and traditional YOLT network and improved YOLT network structure are respectively such as table 1,2 institute of table
Show:
Table 1
Table 2
The specific steps of training process are as shown in Figure 3:
(B1) ship datebase is sent into deep learning network, chooses the good deep learning mould of ImageNet database training
Type is as basic network model, and the learning rate that training starts is 0.0001, and the parameter steps of percentage regulation learning network comes later
Change learning rate.
(B2) image in ship datebase is zoomed into fixed dimension at random, obtains training image.
The method of image scaling are as follows: image zooms to fixed dimension at random, and from 320 to 608, step-length is the size after scaling
32, it is therefore an objective to allow the high width of the convolution characteristic pattern generated below to be all odd number, can produce a central gridding in this way.Because of observation
It arrives, big target has generally taken up the middle position of image, and the position of these targets can be only predicted with a central gridding, no
It will then be predicted with 4 intermediate grids.
(B3) training image is divided into n*n grid, 7 bounding boxes of each grid forecasting.
The range of n is 10 to 19.Since YOLT uses anchor boxes mechanism, and 5 are used in network
Anchors, this patent is in order to obtain more boundary candidate frames, 7 anchors of improved YOLT Web vector graphic, and
K-means Mean Method is utilized in anchors acquiring size.
(B4) training image is sent into deep learning network, it is defeated after convolutional layer, pond layer, Passthrough layers
Convolution characteristic pattern out.
Here, in order to obtain the more inherent feature of ship target, improved YOLT network is before the last one pond layer
5 convolutional layers and pond layer are increased, but due to after multiple convolution, being easily lost ship target detailed information abundant,
Improved YOLT network before the layer of penultimate pond more increases one passthrough layers, by the richer language of shallow-layer
Adopted information is transmitted backward.
Batch processing normalization operation will be carried out after each layer of convolutional layer, passthrough layers defeated by a upper convolutional layer
Characteristic pattern out and the characteristic pattern of the convolutional layer output in shallow-layer network are overlapped, and carry out normalization operation, are obtained final
Characteristic pattern thinning method it is as follows:
B41: after training image is sent into network, successively pass through the convolutional layer of different convolution kernels, deep learning network volume
The convolution kernel size of lamination is respectively 3*3,1*1, is used alternatingly, and the output of each convolutional layer will be as next layer of input.
B42: characteristic pattern obtained in step B41 inputs in the layer of pond, compresses to characteristic pattern size, pond method is
Average value pond, compressed dimension are chosen for 2 times of windows.
B43: image to be detected is input in deep learning network, is repeated above-mentioned B41, B42 step to characteristic pattern size and is
n*n。
B44: the feature that characteristic pattern obtained in step B43 is exported with passthrough layers with convolutional layer in shallow-layer network
Figure size is that (2*n) * (2*n) is connected, and obtains one in one convolution operation of progress later and has shallow semantic information
N*n characteristic pattern.The loss of Small object information after carrying out convolution, pondization operation in order to prevent, obtains shallow-layer language more abundant
Adopted information can add passthrough layers, backward by more shallow semantic information after the layer of penultimate pond here
It passes.
Step B45: the characteristic pattern of the n*n size in step B44 is inputted into convolutional layer.
(B5) it is predicted with the convolution characteristic pattern of output using convolution, sliding window sampling is carried out on convolution characteristic pattern, according to
Step (B3), 7 different size of bounding boxes of each grid forecasting eventually predict n*n*7 bounding box, each bounding box
It returns position, confidence level and classification information and weight is adjusted by backpropagation then according to calculated loss function value.
The improvement of the calculating of loss function value: due to finding averagely there is the value of target weight value avg_obj to deposit in training
In larger difference, such as when 16 images of primary training, when this 16 being averaged for images return have target weight value avg_obj big
When partially larger than 0.5, several average accuracies for averagely having target weight value avg_obj value relatively low, finally training always are had
It can be because these relatively low be averaged target weight value avg_obj and drag down.Therefore one threshold value a=0.5 is set, when flat
When thering is target weight value avg_obj to be less than threshold value a, lossnew=lossold/ (avg_obj+0.001), otherwise lossnew=
lossold, wherein lossoldIt is the loss function of former YOLT network, lossnewIt is the loss function of improved YOLT network.Threshold
Value a can be properly increased according to training.Invention increases the losses that those are difficult to the image detected, in order to improve those
Relatively low being averaged has target weight value avg_obj, to improve the average accuracy of ship detecting.Avg_obj adds 0.001
Purpose loss when being to prevent from starting to trainnewTend to just infinite.
(B6) terminate to train when reaching maximum number of iterations or when loss function value has reached convergence state, obtain final
Deep learning network and weight file for ship detecting.
Three, it detection process: is detected in unmanned plane image using trained deep learning network and weight file
Ship target, and output test result.Its specific steps is as shown in Figure 4:
(C1) Aerial Images to be detected are sent into trained deep learning network, will be taken photo by plane figure according to step (B2)
As zooming to fixed dimension, step step (B3), (B4) are repeated, final characteristic pattern is exported.
In the present embodiment, in order to obtain fair amount boundary candidate frame number, scale the images to 416*416
When, 13*13*7 bounding box is obtained, both ensure that the accuracy of detection will not increase the burden of detection speed in this way.
Shown in Aerial Images to be detected such as Fig. 5 (civilian ship) and Fig. 6 (military boats and ships).
(C2) it is predicted with final characteristic pattern using convolution, sliding window sampling is carried out on convolution characteristic pattern, eventually in advance
N*n*7 bounding box is surveyed, each bounding box returns position, confidence level and classification information,
(C3) threshold value is set, the low bounding box of score is filtered out, non-maximum is carried out to the bounding box remained
Inhibition processing, obtains final testing result by classifier later, to realize the accurate positioning and identification of ship.
The method for obtaining final ship detecting result: non-maxima suppression handles (NMS), will after removing redundancy window
Output valve is input in Softmax classifier, obtains maximum class probability;According to maximum class probability value, ship detecting is obtained.
Shown in final ship detecting result such as Fig. 7 (civilian ship) and Fig. 8 (military boats and ships).
It proves after tested, the present invention can reach 0.923 to the accuracy rate of ship detecting, wherein the accuracy rate of military ship
It is 0.931, the accuracy rate of civilian ship is 0.915, it is wider for different types of ship detecting adaptability, for due to nobody
Machine take photo by plane shoot reason generation distortion equally there is preferable effect, the detection suitable for multiple ships.
Embodiment is merely illustrative of the invention's technical idea, and this does not limit the scope of protection of the present invention, it is all according to
Technical idea proposed by the present invention, any changes made on the basis of the technical scheme are fallen within the scope of the present invention.
Claims (10)
1. a kind of ship detecting method in unmanned plane image based on deep learning, which is characterized in that including following step
It is rapid:
(1) acquisition includes the unmanned plane image of military boats and ships and civilian ship and is labeled, and obtains ship datebase;
(2) ship datebase that step (1) obtains is sent into deep learning network to be trained, until network convergence;
(3) ship in unmanned plane image is detected using the trained deep learning network of step (2) and weight file
Target, and output test result.
2. the ship detecting method in the unmanned plane image based on deep learning, feature exist according to claim 1
In specific step is as follows for step (1):
(A1) the unmanned plane image comprising ship for training deep learning network is acquired;
(A2) acquired image is subjected to data prediction;
(A3) pretreated image is labeled with rectangle frame, obtains included ship in the coordinate and rectangle frame of rectangle frame
The type of oceangoing ship, finally obtains ship datebase.
3. the ship detecting method in the unmanned plane image based on deep learning, feature exist according to claim 2
In in step (A2), the data prediction includes losing image not comprising ship target and ship target is shown not
To the image of half.
4. the ship detecting method in the unmanned plane image based on deep learning, feature exist according to claim 2
In, need to mark two class data: military boats and ships and civilian ship in step (A3), first using real-coded GA format record
The location information of rectangle frame is then converted into the data format that improved deep learning network needs.
5. the ship detecting method in the unmanned plane image based on deep learning, feature exist according to claim 1
In deep learning network described in step (2) is improved YOLT network, and the structure of improved YOLT network is as shown in the table:
Training data or test data input network from the 0th layer of convolutional layer, successively pass through 0,1,2 ..., 28 layers of processing, finally from
28th layer of convolutional layer output;Wherein, at the 12nd layer Passthrough layers, the 6th layer of convolutional layer has been connect with 11th layer pond layer
Come, is then input to the 13rd layer of convolutional layer;At the 26th layer Passthrough layers, by the 17th layer of convolutional layer and the 25th layer of convolutional layer
It connects, is then input to the 27th layer of convolutional layer.
6. the ship detecting method in the unmanned plane image based on deep learning, feature exist according to claim 5
In the improved YOLT network uses anchor boxes mechanism, using 7 anchors, and utilizes the mean value side k-means
Method obtains anchors size.
7. the ship detecting method in the unmanned plane image based on deep learning, feature exist according to claim 1
In specific step is as follows for step (2):
(B1) ship datebase is sent into deep learning network, chooses the good deep learning model of ImageNet database training and makees
For basic network model, the learning rate that training starts is 0.0001, and the parameter steps of percentage regulation learning network changes later
Learning rate;
(B2) image in ship datebase is zoomed into fixed dimension at random, obtains training image;
(B3) training image is divided into n*n grid, 7 bounding boxes of each grid forecasting;
(B4) training image is sent into deep learning network, volume is exported after convolutional layer, pond layer, Passthrough layers
Product characteristic pattern;
(B5) it is predicted with the convolution characteristic pattern of output using convolution, sliding window sampling is carried out on convolution characteristic pattern, according to step
(B3), 7 different size of bounding boxes of each grid forecasting eventually predict that n*n*7 bounding box, each bounding box return
Position, confidence level and classification information adjust weight by backpropagation then according to calculated loss function value;
(B6) terminate to train when reaching maximum number of iterations or when loss function value has reached convergence state, obtain eventually for
The deep learning network and weight file of ship detecting.
8. the ship detecting method in the unmanned plane image based on deep learning, feature exist according to claim 7
In in step (B2), zooming to fixed dimension M*M, M ∈ [320,608], and step-length is 32;In step (B3), n ∈
[10,19]。
9. the ship detecting method in the unmanned plane image based on deep learning, feature exist according to claim 7
In, in step (B5), loss function to recalculate method as follows:
One threshold value a is set, when averagely there is target weight value avg_obj to be less than threshold value a, lossnew=lossold/ (avg_
Obj+0.001), otherwise lossnew=lossold, wherein lossoldIt is the loss function of former YOLT network, lossnewIt is to improve
YOLT network loss function.
10. the ship detecting method in the unmanned plane image based on deep learning, feature exist according to claim 7
In specific step is as follows for step (3):
(C1) Aerial Images to be detected are sent into trained deep learning network, Aerial Images contracts according to step (B2)
It is put into fixed dimension, step step (B3), (B4) is repeated, exports final characteristic pattern;
(C2) it is predicted with final characteristic pattern using convolution, sliding window sampling is carried out on convolution characteristic pattern, eventually predicts n*
N*7 bounding box, each bounding box return position, confidence level and classification information;
(C3) threshold value is set, the low bounding box of score is filtered out, non-maxima suppression is carried out to the bounding box remained
Processing, obtains final testing result by classifier later, to realize the accurate positioning and identification of ship.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811330500.6A CN109657541A (en) | 2018-11-09 | 2018-11-09 | A kind of ship detecting method in unmanned plane image based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811330500.6A CN109657541A (en) | 2018-11-09 | 2018-11-09 | A kind of ship detecting method in unmanned plane image based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109657541A true CN109657541A (en) | 2019-04-19 |
Family
ID=66110767
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811330500.6A Pending CN109657541A (en) | 2018-11-09 | 2018-11-09 | A kind of ship detecting method in unmanned plane image based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109657541A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110070142A (en) * | 2019-04-29 | 2019-07-30 | 上海大学 | A kind of marine vessel object detection method based on YOLO neural network |
CN110175524A (en) * | 2019-04-26 | 2019-08-27 | 南京航空航天大学 | A kind of quick vehicle checking method of accurately taking photo by plane based on lightweight depth convolutional network |
CN110321811A (en) * | 2019-06-17 | 2019-10-11 | 中国工程物理研究院电子工程研究所 | Depth is against the object detection method in the unmanned plane video of intensified learning |
CN111476159A (en) * | 2020-04-07 | 2020-07-31 | 哈尔滨工业大学 | Method and device for training and detecting detection model based on double-angle regression |
CN111611918A (en) * | 2020-05-20 | 2020-09-01 | 重庆大学 | Traffic flow data set acquisition and construction method based on aerial photography data and deep learning |
CN111652321A (en) * | 2020-06-10 | 2020-09-11 | 江苏科技大学 | Offshore ship detection method based on improved YOLOV3 algorithm |
CN111950523A (en) * | 2020-08-28 | 2020-11-17 | 珠海大横琴科技发展有限公司 | Ship detection optimization method and device based on aerial photography, electronic equipment and medium |
CN112037582A (en) * | 2020-09-16 | 2020-12-04 | 浙江工业大学 | Cross-sea bridge ship collision early warning method based on risk prediction |
CN112149567A (en) * | 2020-09-23 | 2020-12-29 | 中国科学院长春光学精密机械与物理研究所 | Offshore target detection method, device and equipment |
CN112164097A (en) * | 2020-10-20 | 2021-01-01 | 南京莱斯网信技术研究院有限公司 | Ship video detection sample acquisition method |
CN112416968A (en) * | 2020-12-09 | 2021-02-26 | 中国船舶工业系统工程研究院 | Unmanned ship data management system supporting data set generation |
CN112766221A (en) * | 2021-02-01 | 2021-05-07 | 福州大学 | Ship direction and position multitask-based SAR image ship target detection method |
CN112819068A (en) * | 2021-01-29 | 2021-05-18 | 南京长江油运有限公司 | Deep learning-based real-time detection method for ship operation violation behaviors |
CN113361503A (en) * | 2021-08-10 | 2021-09-07 | 江苏久智环境科技服务有限公司 | Intelligent garden arbor quantity detection method and system based on unmanned aerial vehicle aerial photography |
WO2022252892A1 (en) * | 2021-06-02 | 2022-12-08 | Ping An Technology (Shenzhen) Co., Ltd. | System and method for image-based crop identification |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107451602A (en) * | 2017-07-06 | 2017-12-08 | 浙江工业大学 | A kind of fruits and vegetables detection method based on deep learning |
CN107563303A (en) * | 2017-08-09 | 2018-01-09 | 中国科学院大学 | A kind of robustness Ship Target Detection method based on deep learning |
WO2018067080A1 (en) * | 2016-10-07 | 2018-04-12 | Aselsan Elektronik Sanayi Ve Ticaret Anonim Sirketi | A marine vessel identification method |
CN108009525A (en) * | 2017-12-25 | 2018-05-08 | 北京航空航天大学 | A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks |
CN108614996A (en) * | 2018-04-04 | 2018-10-02 | 杭州电子科技大学 | A kind of military ships based on deep learning, civilian boat automatic identifying method |
CN108960143A (en) * | 2018-07-04 | 2018-12-07 | 北京航空航天大学 | Detect deep learning method in a kind of naval vessel in High Resolution Visible Light remote sensing images |
-
2018
- 2018-11-09 CN CN201811330500.6A patent/CN109657541A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018067080A1 (en) * | 2016-10-07 | 2018-04-12 | Aselsan Elektronik Sanayi Ve Ticaret Anonim Sirketi | A marine vessel identification method |
CN107451602A (en) * | 2017-07-06 | 2017-12-08 | 浙江工业大学 | A kind of fruits and vegetables detection method based on deep learning |
CN107563303A (en) * | 2017-08-09 | 2018-01-09 | 中国科学院大学 | A kind of robustness Ship Target Detection method based on deep learning |
CN108009525A (en) * | 2017-12-25 | 2018-05-08 | 北京航空航天大学 | A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks |
CN108614996A (en) * | 2018-04-04 | 2018-10-02 | 杭州电子科技大学 | A kind of military ships based on deep learning, civilian boat automatic identifying method |
CN108960143A (en) * | 2018-07-04 | 2018-12-07 | 北京航空航天大学 | Detect deep learning method in a kind of naval vessel in High Resolution Visible Light remote sensing images |
Non-Patent Citations (1)
Title |
---|
程平: "无人机遥感图像海上舰船目标识别技术的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110175524A (en) * | 2019-04-26 | 2019-08-27 | 南京航空航天大学 | A kind of quick vehicle checking method of accurately taking photo by plane based on lightweight depth convolutional network |
CN110070142A (en) * | 2019-04-29 | 2019-07-30 | 上海大学 | A kind of marine vessel object detection method based on YOLO neural network |
CN110321811A (en) * | 2019-06-17 | 2019-10-11 | 中国工程物理研究院电子工程研究所 | Depth is against the object detection method in the unmanned plane video of intensified learning |
CN110321811B (en) * | 2019-06-17 | 2023-05-02 | 中国工程物理研究院电子工程研究所 | Target detection method in unmanned aerial vehicle aerial video for deep reverse reinforcement learning |
CN111476159A (en) * | 2020-04-07 | 2020-07-31 | 哈尔滨工业大学 | Method and device for training and detecting detection model based on double-angle regression |
CN111476159B (en) * | 2020-04-07 | 2023-04-07 | 哈尔滨工业大学 | Method and device for training and detecting detection model based on double-angle regression |
CN111611918A (en) * | 2020-05-20 | 2020-09-01 | 重庆大学 | Traffic flow data set acquisition and construction method based on aerial photography data and deep learning |
CN111611918B (en) * | 2020-05-20 | 2023-07-21 | 重庆大学 | Traffic flow data set acquisition and construction method based on aerial data and deep learning |
CN111652321A (en) * | 2020-06-10 | 2020-09-11 | 江苏科技大学 | Offshore ship detection method based on improved YOLOV3 algorithm |
CN111652321B (en) * | 2020-06-10 | 2023-06-02 | 江苏科技大学 | Marine ship detection method based on improved YOLOV3 algorithm |
CN111950523A (en) * | 2020-08-28 | 2020-11-17 | 珠海大横琴科技发展有限公司 | Ship detection optimization method and device based on aerial photography, electronic equipment and medium |
CN112037582A (en) * | 2020-09-16 | 2020-12-04 | 浙江工业大学 | Cross-sea bridge ship collision early warning method based on risk prediction |
CN112149567A (en) * | 2020-09-23 | 2020-12-29 | 中国科学院长春光学精密机械与物理研究所 | Offshore target detection method, device and equipment |
CN112164097B (en) * | 2020-10-20 | 2024-03-29 | 南京莱斯网信技术研究院有限公司 | Ship video detection sample collection method |
CN112164097A (en) * | 2020-10-20 | 2021-01-01 | 南京莱斯网信技术研究院有限公司 | Ship video detection sample acquisition method |
CN112416968A (en) * | 2020-12-09 | 2021-02-26 | 中国船舶工业系统工程研究院 | Unmanned ship data management system supporting data set generation |
CN112819068A (en) * | 2021-01-29 | 2021-05-18 | 南京长江油运有限公司 | Deep learning-based real-time detection method for ship operation violation behaviors |
CN112819068B (en) * | 2021-01-29 | 2023-09-26 | 南京长江油运有限公司 | Ship operation violation behavior real-time detection method based on deep learning |
CN112766221B (en) * | 2021-02-01 | 2022-06-14 | 福州大学 | Ship direction and position multitasking-based SAR image ship target detection method |
CN112766221A (en) * | 2021-02-01 | 2021-05-07 | 福州大学 | Ship direction and position multitask-based SAR image ship target detection method |
WO2022252892A1 (en) * | 2021-06-02 | 2022-12-08 | Ping An Technology (Shenzhen) Co., Ltd. | System and method for image-based crop identification |
CN113361503A (en) * | 2021-08-10 | 2021-09-07 | 江苏久智环境科技服务有限公司 | Intelligent garden arbor quantity detection method and system based on unmanned aerial vehicle aerial photography |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109657541A (en) | A kind of ship detecting method in unmanned plane image based on deep learning | |
CN107818326B (en) | A kind of ship detection method and system based on scene multidimensional characteristic | |
CN114202696B (en) | SAR target detection method and device based on context vision and storage medium | |
CN110378308B (en) | Improved port SAR image near-shore ship detection method based on fast R-CNN | |
CN109241913A (en) | In conjunction with the ship detection method and system of conspicuousness detection and deep learning | |
CN112766087A (en) | Optical remote sensing image ship detection method based on knowledge distillation | |
CN108171752A (en) | A kind of sea ship video detection and tracking based on deep learning | |
CN111723748A (en) | Infrared remote sensing image ship detection method | |
CN109784278B (en) | Deep learning-based marine small and weak motion ship real-time detection method | |
CN109271856A (en) | Remote sensing image object detection method based on expansion residual error convolution | |
CN105022990A (en) | Water surface target rapid-detection method based on unmanned vessel application | |
CN110458160A (en) | A kind of unmanned boat waterborne target recognizer based on depth-compression neural network | |
CN108957453A (en) | A kind of high-precision pre-filter method and recognition methods based on multiple target tracking | |
CN110516605A (en) | Any direction Ship Target Detection method based on cascade neural network | |
CN108830185B (en) | Behavior identification and positioning method based on multi-task joint learning | |
CN109682378A (en) | A kind of unmanned plane indoor positioning and multi-object tracking method based entirely on visual information | |
CN110472500A (en) | A kind of water surface sensation target fast algorithm of detecting based on high speed unmanned boat | |
CN112149591A (en) | SSD-AEFF automatic bridge detection method and system for SAR image | |
CN109063549A (en) | High-resolution based on deep neural network is taken photo by plane video moving object detection method | |
CN106372590A (en) | Sea surface ship intelligent tracking system and method based on machine vision | |
Shaodan et al. | A ship target location and mask generation algorithms base on Mask RCNN | |
Zhang Sr et al. | A ship target tracking algorithm based on deep learning and multiple features | |
Zhang et al. | Nearshore vessel detection based on Scene-mask R-CNN in remote sensing image | |
Huang et al. | Identity recognition on waterways: A novel ship information tracking method based on multimodal data | |
Liu et al. | Target detection and tracking algorithm based on improved Mask RCNN and LMB |
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 |
Application publication date: 20190419 |
|
RJ01 | Rejection of invention patent application after publication |