CN109800714A - A kind of ship detecting system and method based on artificial intelligence - Google Patents
A kind of ship detecting system and method based on artificial intelligence Download PDFInfo
- Publication number
- CN109800714A CN109800714A CN201910058814.3A CN201910058814A CN109800714A CN 109800714 A CN109800714 A CN 109800714A CN 201910058814 A CN201910058814 A CN 201910058814A CN 109800714 A CN109800714 A CN 109800714A
- Authority
- CN
- China
- Prior art keywords
- faster
- ship
- module
- model
- artificial intelligence
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The ship detecting system based on artificial intelligence that the invention discloses a kind of, including video frame extraction and transcoding module, Faster R-CNN model, vessel motions feature calculation module and the model calculation scheduler module based on ResNet68 of pre-training, the video frame extraction obtains the image in video with transcoding module and transcoding carries out subsequent operation for model calculation scheduler module;The pre-training model carries out target detection and classification to single-frame images, exports candidate frame, classification results, confidence score;The vessel motions result computing module further screens the testing result of Faster R-CNN model and calculates ship speed, the direction of motion;The model calculation scheduler module is distributed each Process Synchronization and is calculated using Faster R-CNN model.The present invention has the characteristics that transplanting is simple, accuracy rate is high, adapts to extreme weather conditions, detectable channel span is big.
Description
Technical field
The invention belongs to artificial intelligence technology applications, and in particular to a kind of ship detecting system and side based on artificial intelligence
Method.
Background technique
Object identification is widely used in intelligent monitoring device instantly, such as more mature recognition of face, license plate
Identification etc..But the vehicle identification relative to fixed lane, the ship identification on navigation channel is with regard to complex.Its difficult point essentially consists in
River channel is not fixed, ship appearance difference is big, and background interference more (vegetation reflective, on the bank of such as water surface), detection range are wide
(river surface detection range is wide up to 200 meters or more), different navigation channels environment are different, this results in traditional image detection algorithm to be difficult pair
This provides a pervasive method.Image detection algorithm based on artificial intelligence overcomes this problem well.
In recent years, artificial intelligence has prominent achievement in graph image field, with R-CNN(region with CNN
Features proposition), starts to be widely used in now NI Vision Builder for Automated Inspection, including object identification, automatic segmentation, nobody drives
Sail equal fields.Pass through selective search after classifying in advance and use CNNs(large size convolutional neural networks) feature is extracted, it can be very
The target object of good detection different scale, different length-width ratios, solves the difficult point that ship appearance difference is big in ship detection.Together
When also have high discrimination for fuzzy pictures caused by the ship or rain and snow weather that are blocked.Wherein using RPN nerve
Network (Region Proposal Network) extracts data of the couple candidate detection frame as CNNs, can greatly improve processing
Speed.This method is known as Faster R-CNN.
In addition, the depth of network has been largely fixed the performance of model, more network numbers of plies are more complicated to extract
Feature mode create possibility so that model is more outstanding.But classical network will appear network degeneration after intensification
Problem tends to saturation or degeneration as depth increases network accuracy.Thus in a model, ResNet(depth residual error is used
Network) the common VGG network of substitution, solve the degenerate problem of depth network while network is deepened by residual error study
Ensure the accuracy rate of model.Faster R-CNN based on ResNet68, it is more common based on VGG16 model accuracy rate more
Height can be better used in the ship video detection of different shape sizes.
To sum up, for the different ships in different navigation channels, monitoring is regarded using the Faster R-CNN model after specialized training
Frequency carries out object identification, has the advantages such as applied widely, accuracy rate is high relative to traditional object identification method, while extreme
Situation (such as rain, snow, remote) has descended extraordinary identification performance.Even for some special monitoring points, also only need
It will be to model special training, without redesigning overall plan.On the basis of the recognition result, using tracing algorithm,
It obtains and often searches the information such as speed and driving direction by the monitoring point ship.By calculating GPU the scheduling of power, so that multiple moulds
Type can be calculated simultaneously, to improve calculating speed, to match ship average running speed, accomplish to identify in time anti-in time
Feedback.
Summary of the invention
It is general the purpose of the present invention is being difficult to propose the complex situations of ship detection in navigation channel for traditional graph method
The problem of suitable highly effective algorithm, propose a kind of ship detecting method based on artificial intelligence.The present invention makes full use of CNNs network to exist
Advantage in feature extraction, and be aided with RPN neural network and provide candidate region to shorten the calculating time, it is replaced and is passed with ResNet68
VGG16 network of uniting increases network depth and improves network accuracy rate, obtains ship using tracing algorithm finally for the target detected
Speed, the information such as direction of navigation.Its specific technical solution is as follows:
A kind of ship detecting system based on artificial intelligence, including video frame extraction and transcoding module, pre-training based on
Faster R-CNN model, vessel motions feature calculation module and the model calculation scheduler module of ResNet68, the video frame
It extracts with the image in transcoding module acquisition video and transcoding is for the progress subsequent operation of model calculation scheduler module;The pre-training
The Faster R-CNN model based on ResNet68 target detection and classification, output candidate frame, classification are carried out to single-frame images
As a result, confidence score;The vessel motions result computing module carries out into one the testing result of Faster R-CNN model
Step screens and calculates ship speed, the direction of motion;The model calculation scheduler module distributes each Process Synchronization and uses Faster
R-CNN model is calculated.
Further, the video frame extraction and transcoding module receive the video frame picture that front end camera is transmitted back to by network
Face calls GPU operation, single frames picture is quickly converted into BMP format by YUV.
Further, the Faster R-CNN model based on ResNet68 of the pre-training, using being based on
The Faster R-CNN model of ResNet68 amplifies the small river ship figure for 1024x600 resolution ratio using handmarking's retraction
As being trained.
Further, the vessel motions feature calculation module is examined by comparing previous frame testing result and present frame
The value of the confidence for surveying result is further screened, and calculates vessel motions speed and direction by orientation of the ship on photo.
Further, Faster R-CNN model is separately operable and works as in subprocess by the model calculation scheduler module
In, process, which is distributed, according to GPU current idle situation carries out operational model.
A kind of ship detecting method based on artificial intelligence, includes the following steps:
One, the image that front end surveillance device obtains passes through network transmission;
Two, described image is transcoded into single-frame images;
Three, the Faster R-CNN model based on ResNet68 of pre-training in subprocess is selected to carry out detection identification;
Four, it combines previous content frame further to screen recognition result and provides vessel motions characteristic.
It is difficult to ship recognition detection in river ship video present invention is mainly used for current traditional graph algorithm is solved
The problem of, and have the characteristics that simple transplanting, accuracy rate height, adaptation extreme weather conditions, detectable channel span are big.Due to this
Reasonable GPU distribution method scheduling model operation is used in the invention technology, greatly improves the speed of intelligent algorithm
Degree, allows to obtain the testing result of single frame video in time.For these reasons, according to system of the present invention, Ke Yi
Under prediction based on Faster R-CNN model, the ship in the monitor video of navigation channel is effectively identified.
Detailed description of the invention
Fig. 1 is flow chart of the invention;
Fig. 2 is the structure block schematic diagram of residual error network;
Fig. 3 is ResNet68 residual error network.
Specific embodiment
The embodiment of the invention will now be described in detail with reference to the accompanying drawings.
As shown in Figure 1, the ship detecting method of the invention based on artificial intelligence, the system of use includes video frame extraction
With Faster R-CNN model, vessel motions feature calculation module and the model based on ResNet68 of transcoding module, pre-training
Operation scheduler module.The image that front end surveillance device obtains by network transmission to the video frame extraction and transcoding module,
By model calculation after the video frame extraction and transcoding module transcoding single-frame images
The Faster R-CNN model based on ResNet68 of pre-training carries out detection identification in scheduler module selection subprocess, and
Result is fed back to vessel motions feature calculation module further to screen in conjunction with previous content frame and provide vessel motions characteristic
According to.
Above-mentioned video frame extraction and transcoding module, including obtain video from network and extract a frame out, by GPU by the frame
It is RGB, input model operation scheduler module by YUV transcoding.
Above-mentioned model calculation scheduler module inquires current subprocess situation after receiving data, if current available free son into
The picture is sent model to detect by Cheng Ze, otherwise abandons the frame.
The Faster R-CNN model based on ResNet68 of above-mentioned pre-training is small using the amplification of handmarking's retraction to be
The river ship image of 1024x600 resolution ratio is trained.
As shown in Fig. 2, solving degenerate problem by introducing Resnet101 depth residual error learning framework.In form, clear
Ground allows the mapping of these layer of regression criterion, it would be desirable to base map be expressed as H(x), the non-linear layer of stacking is fitted another
Map F(x): H (x)-x.Original mappings are rewritten as F (x)+x.Formula F (x)+x passes through with the feedforward neural network fast connected
To realize.Quick connection is to skip the connection of one or more layers, and quick connection simply executes identical mapping, and outputs it
It is added to the output of stack layer.
Residual error network based on above-mentioned ResNet can be used directly when outputting and inputting dimension having the same
Identical quick connection considers two options when dimension increases: (1) quick connection still carries out identical mapping, extra padding zero
Input is to increase dimension.(2) the projection connection in equation is used for matching dimensionality (being completed by 1x1 convolution).Fig. 3 is building
Resnet68 residual error network.
Above-mentioned vessel motions feature calculation module is further screened by the detection case of comparison present frame and previous frame
Testing result and calculating vessel motions feature.Should occur (driving into imaging region for the first time in 1/4 position in the visual field for ship
Domain) or there is (sailing out of imaging area) for the last time, ship in addition to this is recorded as judging by accident.By judging in successive two frame
The motion feature of ship determines the ship of ship and previous frame in present frame in the overlapping cases and picture of the detection block of ship
It whether is same ship.By the relative distance of same ship in two frames, travel speed and the direction of ship are updated.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations;To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement;And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (6)
1. a kind of ship detecting system based on artificial intelligence, including video frame extraction and transcoding module, pre-training based on
Faster R-CNN model, vessel motions feature calculation module and the model calculation scheduler module of ResNet68, it is characterised in that:
The video frame extraction and transcoding module obtain the image in video and transcoding carries out subsequent operation for model calculation scheduler module;
The Faster R-CNN model based on ResNet68 of the pre-training carries out target detection and classification to single-frame images, and output is waited
Select frame, classification results, confidence score;Testing result of the vessel motions result computing module to Faster R-CNN model
It is further screened and calculates ship speed, the direction of motion;The model calculation scheduler module, which distributes each Process Synchronization, to be made
It is calculated with Faster R-CNN model.
2. a kind of ship detecting system based on artificial intelligence as described in claim 1, it is characterised in that the video frame mentions
It takes and the video frame picture that front end camera is transmitted back to is received by network with transcoding module, GPU operation is called, quickly by single frames picture
BMP format is converted by YUV.
3. a kind of ship detecting system based on artificial intelligence as described in claim 1, it is characterised in that the pre-training
The Faster R-CNN model based on ResNet68, using the Faster R-CNN model based on ResNet68, using artificial
Label retraction is amplified small to be trained for the river ship image of 1024x600 resolution ratio.
4. a kind of ship detecting system based on artificial intelligence as described in claim 1, it is characterised in that the ship fortune
Dynamic feature calculation module, the value of the confidence by comparing previous frame testing result and present frame testing result is further screened, and is led to
It crosses orientation of the ship on photo and calculates vessel motions speed and direction.
5. a kind of ship detecting system based on artificial intelligence as described in claim 1, it is characterised in that the model fortune
Scheduler module is calculated, Faster R-CNN model is separately operable in subprocess, distributes process according to GPU current idle situation
Carry out operational model.
6. a kind of ship detecting method based on artificial intelligence, includes the following steps:
One, the image that front end surveillance device obtains passes through network transmission;
Two, described image is transcoded into single-frame images;
Three, the Faster R-CNN model based on ResNet68 of pre-training in subprocess is selected to carry out detection identification;
Four, it combines previous content frame further to screen recognition result and provides vessel motions characteristic.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910058814.3A CN109800714A (en) | 2019-01-22 | 2019-01-22 | A kind of ship detecting system and method based on artificial intelligence |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910058814.3A CN109800714A (en) | 2019-01-22 | 2019-01-22 | A kind of ship detecting system and method based on artificial intelligence |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109800714A true CN109800714A (en) | 2019-05-24 |
Family
ID=66559961
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910058814.3A Pending CN109800714A (en) | 2019-01-22 | 2019-01-22 | A kind of ship detecting system and method based on artificial intelligence |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109800714A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110619653A (en) * | 2019-08-26 | 2019-12-27 | 衢州市港航管理局 | Early warning control system and method for preventing collision between ship and bridge based on artificial intelligence |
CN110991230A (en) * | 2019-10-25 | 2020-04-10 | 湖北富瑞尔科技有限公司 | Method and system for detecting ships by remote sensing images in any direction based on rotating candidate frame |
CN111523558A (en) * | 2019-12-27 | 2020-08-11 | 珠海大横琴科技发展有限公司 | Ship shielding detection method and device based on electronic purse net and electronic equipment |
CN113095325A (en) * | 2021-05-11 | 2021-07-09 | 浙江华是科技股份有限公司 | Ship identification method and device and computer readable storage medium |
CN113592799A (en) * | 2021-07-22 | 2021-11-02 | 象山电力实业有限公司 | Protection method and system for electric power submarine cable |
CN113628208A (en) * | 2021-08-30 | 2021-11-09 | 北京中星天视科技有限公司 | Ship detection method, device, electronic equipment and computer readable medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101022541A (en) * | 2006-02-13 | 2007-08-22 | 中兴通讯股份有限公司 | Image monitoring system and method under server/customer end constitution |
CN105187721A (en) * | 2015-08-31 | 2015-12-23 | 广州市幸福网络技术有限公司 | An identification camera and method for rapidly extracting portrait features |
CN108055501A (en) * | 2017-11-22 | 2018-05-18 | 天津市亚安科技有限公司 | A kind of target detection and the video monitoring system and method for tracking |
CN108830172A (en) * | 2018-05-24 | 2018-11-16 | 天津大学 | Aircraft remote sensing images detection method based on depth residual error network and SV coding |
-
2019
- 2019-01-22 CN CN201910058814.3A patent/CN109800714A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101022541A (en) * | 2006-02-13 | 2007-08-22 | 中兴通讯股份有限公司 | Image monitoring system and method under server/customer end constitution |
CN105187721A (en) * | 2015-08-31 | 2015-12-23 | 广州市幸福网络技术有限公司 | An identification camera and method for rapidly extracting portrait features |
CN108055501A (en) * | 2017-11-22 | 2018-05-18 | 天津市亚安科技有限公司 | A kind of target detection and the video monitoring system and method for tracking |
CN108830172A (en) * | 2018-05-24 | 2018-11-16 | 天津大学 | Aircraft remote sensing images detection method based on depth residual error network and SV coding |
Non-Patent Citations (2)
Title |
---|
杨路江 等: "基于Faster_R_CNN模型的遥感污水处理厂目标检测", 《第五届高分辨率对地观测学术年会论文集》 * |
马鹏阁 羊毅 著: "《多脉冲激光雷达》", 31 December 2017 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110619653A (en) * | 2019-08-26 | 2019-12-27 | 衢州市港航管理局 | Early warning control system and method for preventing collision between ship and bridge based on artificial intelligence |
CN110991230A (en) * | 2019-10-25 | 2020-04-10 | 湖北富瑞尔科技有限公司 | Method and system for detecting ships by remote sensing images in any direction based on rotating candidate frame |
CN110991230B (en) * | 2019-10-25 | 2023-08-11 | 湖北富瑞尔科技有限公司 | Method and system for detecting ship by using remote sensing images in any direction based on rotation candidate frame |
CN111523558A (en) * | 2019-12-27 | 2020-08-11 | 珠海大横琴科技发展有限公司 | Ship shielding detection method and device based on electronic purse net and electronic equipment |
CN113095325A (en) * | 2021-05-11 | 2021-07-09 | 浙江华是科技股份有限公司 | Ship identification method and device and computer readable storage medium |
CN113592799A (en) * | 2021-07-22 | 2021-11-02 | 象山电力实业有限公司 | Protection method and system for electric power submarine cable |
CN113628208A (en) * | 2021-08-30 | 2021-11-09 | 北京中星天视科技有限公司 | Ship detection method, device, electronic equipment and computer readable medium |
CN113628208B (en) * | 2021-08-30 | 2024-02-06 | 北京中星天视科技有限公司 | Ship detection method, device, electronic equipment and computer readable medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109800714A (en) | A kind of ship detecting system and method based on artificial intelligence | |
WO2020173226A1 (en) | Spatial-temporal behavior detection method | |
CN109977812B (en) | Vehicle-mounted video target detection method based on deep learning | |
CN109902600B (en) | Road area detection method | |
CN108009518A (en) | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks | |
CN110263706A (en) | A kind of haze weather Vehicular video Detection dynamic target and know method for distinguishing | |
CN112395951B (en) | Complex scene-oriented domain-adaptive traffic target detection and identification method | |
CN111008632B (en) | License plate character segmentation method based on deep learning | |
CN110705412A (en) | Video target detection method based on motion history image | |
CN117253154B (en) | Container weak and small serial number target detection and identification method based on deep learning | |
CN113888754B (en) | Vehicle multi-attribute identification method based on radar vision fusion | |
CN112149620A (en) | Method for constructing natural scene character region detection model based on no anchor point | |
CN112766056B (en) | Method and device for detecting lane lines in low-light environment based on deep neural network | |
CN113011338B (en) | Lane line detection method and system | |
CN111027511A (en) | Remote sensing image ship detection method based on region of interest block extraction | |
Xiao et al. | Real-time object detection algorithm of autonomous vehicles based on improved yolov5s | |
CN111291684A (en) | Ship board detection method in natural scene | |
CN113033315A (en) | Rare earth mining high-resolution image identification and positioning method | |
CN113706523A (en) | Method for monitoring belt deviation and abnormal operation state based on artificial intelligence technology | |
CN110837769B (en) | Image processing and deep learning embedded far infrared pedestrian detection method | |
CN116246059A (en) | Vehicle target recognition method based on improved YOLO multi-scale detection | |
CN115272850A (en) | Railway wagon BAB type brake adjuster pull rod head breaking fault identification method | |
CN110619653A (en) | Early warning control system and method for preventing collision between ship and bridge based on artificial intelligence | |
CN115166717A (en) | Lightweight target tracking method integrating millimeter wave radar and monocular camera | |
CN113139497B (en) | System and method for identifying object on water surface and application based on 5G MEC |
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: 20190524 |