Skip to main content
Have a personal or library account? Click to login
Improved Faster R-CNN Algorithm for Sea Object Detection Under Complex Sea Conditions Cover

Improved Faster R-CNN Algorithm for Sea Object Detection Under Complex Sea Conditions

By: Liu Yabin,  Yu Jun and  Hu Zhiyi  
Open Access
|Jul 2020

Figures & Tables

Figure 1.

Fast R-CNN model framework

Figure 2.

Region Proposal Networks

Figure 3.

mAP of different K values

Figure 4.

Is a schematic diagram of the shortcomings of The NMS algorithm

Figure 5.

Improved object detection network

Figure 6.

Test results

VALUES OF MAIN NETWORK PARAMETERS

ParametersValuesParametersValues
LEARNING_RATEle-3Bateh_size256
Anchor_Scales[8,16,32]Anchor_RATIOS[0.7, 1.2, 1.3, 2.9]
ITERS85000num_classes6
SOFT_NMS1

COMPARISON OF DETECTION RESULTS OF THREE NETWORK STRUCTURES

Detection methodPassenger_ shipCargo_shipContainer_ shipAircraft_ shipWar_shipmAP
VGG-16 structure50.9%80.6%92.4%98.1%92.8%82.96%
ResNet101 structure54.1%81.0%92.8%99.3%93.0%84.04%
improved ResNet101 structure66.9%82.3%93.65%99.51%93.9%87.25%

ANCHOR_RATIOS FOR DIFFERENT K VALUES

numberK=1K=2K=3K=4K=5K=6K=7
0.61.30.50.70.60.50.5
-1.41.21.21.20.70.6
--1.61.31.31.21.1
results -- 2.91.71.31.2
----2.62.41.8
-----32.1
------2.9
mAP(%) 83.8484.384.3284.9884.2984.5784.34
Time(s) 503510515518524515517
Language: English
Page range: 76 - 82
Published on: Jul 13, 2020
In partnership with: Paradigm Publishing Services
Publication frequency: 4 issues per year

© 2020 Liu Yabin, Yu Jun, Hu Zhiyi, published by Xi’an Technological University
This work is licensed under the Creative Commons Attribution 4.0 License.