A Big Bang-Big Crunch Type-2 Fuzzy Logic System for Explainable Semantic Segmentation of Trees in Satellite Images using HSV Color Space

被引:2
作者
Leon-Garza, Hugo [1 ]
Hagras, Hani [1 ]
Pena-Rios, Anasol [2 ]
Conway, Anthony [2 ]
Owusu, Gilbert [2 ]
机构
[1] Univ Essex, Computat Intelligence Ctr, Sch Comp Sci & Elect Engn, Colchester, Essex, England
[2] BT Plc, BT Labs, Adastral Pk, Ipswich, Suffolk, England
来源
2020 IEEE INTERNATIONAL CONFERENCE ON FUZZY SYSTEMS (FUZZ-IEEE) | 2020年
关键词
Fuzzy Logic System; Neural Networks; explainable AI; semantic segmentation; interpretable models; satellite images; OPTIMIZATION;
D O I
10.1109/fuzz48607.2020.9177611
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
In recent years, new sensor technologies have increased the accessibility of high-resolution satellite images. The information in these images can help to improve activities like urban planning and growth analysis of cities. Additionally, information extracted from these images can be used for taking decisions related to infrastructure planning, e.g. identifying objects that might interfere with network assets like underground cables. To be able to justify the cost of network planning decisions a high degree of interpretability is required. Convolutional Neural Networks (CNNs) are the state of the art for segmenting these images, but like any black box model they do not offer any explanation for their output. In this paper, we present an approach on how to use a Fuzzy Logic System (FLS) for performing explainable semantic segmentation of trees in satellite images. The FLS uses the HSV (hue, saturation, value) of the pixels as inputs and was optimized by using an evolutionary algorithm called Big Bang Big Crunch. The best configuration for the Interval Type-2 FLS has an Intersection over Union metric measure of 60.6%, which is close to the results obtained from neural network, however the proposed FLS provides interpretable outputs which is highly needed for the real-world operation especially in the telecommunication domain.
引用
收藏
页数:7
相关论文
共 32 条
[11]  
Hagras H, 1999, ICRA '99: IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION, VOLS 1-4, PROCEEDINGS, P1005
[12]  
Hagras H., 2009, IEEE SYSTEMS MAN CYB
[13]   Toward Human-Understandable, Explainable AI [J].
Hagras, Hani .
COMPUTER, 2018, 51 (09) :28-36
[14]   Approaches to large-scale urban modeling [J].
Hu, JH ;
You, SY ;
Neumann, U .
IEEE COMPUTER GRAPHICS AND APPLICATIONS, 2003, 23 (06) :62-69
[15]   Semantic Segmentation of Small Objects and Modeling of Uncertainty in Urban Remote Sensing Images Using Deep Convolutional Neural Networks [J].
Kampffmeyer, Michael ;
Salberg, Arnt-Borre ;
Jenssen, Robert .
PROCEEDINGS OF 29TH IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS, (CVPRW 2016), 2016, :680-688
[16]   Big Bang-Big Crunch optimization based interval type-2 fuzzy PID cascade controller design strategy [J].
Kumbasar, Tufan ;
Hagras, Hani .
INFORMATION SCIENCES, 2014, 282 :277-295
[17]  
LYNCH C, 2006, P INT C INF PROC MAN, P1340
[18]  
Maggiori E, 2017, INT GEOSCI REMOTE SE, P3226, DOI 10.1109/IGARSS.2017.8127684
[19]   High-Resolution Aerial Image Labeling With Convolutional Neural Networks [J].
Maggiori, Emmanuel ;
Tarabalka, Yuliya ;
Charpiat, Guillaume ;
Alliez, Pierre .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2017, 55 (12) :7092-7103
[20]  
Mendel J. M., 2017, Uncertain Rule-Based Fuzzy Logic Systems: Introduction and New Directions, VSecond