---
tags: COTAI Research
---

# Airbone Sensor Placement and Data Selection in Geospatial Information Networks/Geospatial Data Transfer

[TOC]

See also our [**research group page**](https://hackmd.io/@COTAI/RemoteSensingResearch)

## Authors
- Thinh Dinh
- Hung Ngo
- Phuong Dao
- [Diep Nguyen](https://www.uts.edu.au/staff/diep.nguyen)

## Intro

Mobile networks are the data highways and, in a fully connected, intelligent digital world, will need to connect everything, including people to vehicles, sensors, data, cloud resources, and even robotic agents, [Giordani et. al](https://ieeexplore.ieee.org/document/9040264). Fifth generation (5G) wireless networks, which are currently being deployed, offer significant advances beyond LTE, but may be unable to meet the full connectivity demands of the future digital society.

The integration of multidimensional networks such as space, air and ground is the future trend of the IoT, [Hong et. al](https://ieeexplore.ieee.org/document/8961915).

This network is critically important for industries such as logistics, mining, agriculture, fisheries, and defense. However, a number of significant technological challenges must be overcome for ISTN through low-cost airborne platforms and high-data-rate backbone links, [Huang et. al](https://ieeexplore.ieee.org/document/8760401).

The large volume of data generated by future connected devices (e.g., sensors in autonomous vehicles) will put a strain on communication technologies, which could not guarantee the required quality of service. It is therefore fundamental to discriminate the value of information to maximize the utility for the end users with (limited) network resources. In this context, machine learning (ML) strategies can evaluate the degree of correlation in observations, or extract features from input vectors and predict the a posteriori probability of a sequence given its entire history [Giordani et. al](https://ieeexplore.ieee.org/document/9040264).

Being a promising paradigm, Mobile Edge Computing (MEC) has been regarded as a key technology-enabler to offer further service innovation and business agility in Space-Air-Ground Network [Huang et. al](https://ieeexplore.ieee.org/document/9048610).

Potential applications of UAV platforms for IoT data collection can be roughly divided into two categories: 1). UAV-enabled IoT networks, in which single or multiple UAV platforms are used to collect data from devices, while terrestrial Base Stations (BSs) are only used for backhaul links or UAV control. 2). UAV-assisted IoT networks , which jointly use UAV platforms and terrestrial BSs to perform data collection [Wang et. al](https://ieeexplore.ieee.org/document/8894454).

For hyperspectral imagery, in [Liao et. al](https://ieeexplore.ieee.org/document/7342891), authors show the plots of number of features and classification accuracy. As can be seen, the accuracy are not monotonically increase as the number of features increases.


This [professor](https://scholar.google.fr/citations?hl=fr&user=6owK2OQAAAAJ&view_op=list_works&sortby=pubdate) need to be investiated. Do a lot on sparse approximation. He just had a magazine about feature extration for hyperspectral images [Rasti et. al.](https://ieeexplore.ieee.org/document/9082155).


To guarantee high quality of monitoring, UAVs should
jointly adjust their coordinates and the orientations of their
cameras to cover objects at or near their frontal view. Some
sensor placement approaches have been presented in previous works both from the theoretical analysis and practical experiments, but they all focus on omnidirectional sensors and haven’t consider the direction of objects. In paper, [Placement of Unmanned Aerial Vehicles for
Directional Coverage in 3D Space](https://ieeexplore.ieee.org/document/9042871), authors defined directional coverage utility to characterize the effectiveness of directional coverage for all objects. Formally, given a fixed number of UAVs and a set of objects in the space, PANDA problem is to deploy the UAVs in the 3D free space, i.e., to determine their coordinates and orientations (the combinations of which we define as arrangements), such that the directional coverage utility for all objects is maximized.

## System Model

- Satelite network/UAV network

## Problem

Feature Selection can impact on the energy consumption of the whole network

Each device generates a set of information. When collecting data, the amount of information impact on the accuracy, system transmission latency, and energy consumption

optimize do phan giai pho va do phan giai khong gian

Denote $s \in S$ where $s$ is a subset of data and $S$ is a set of data, $\theta$ is the location of satelite

Each data owner want to
$$
\max_{s,\theta} Scoring(s,\theta) \\
s.t~~network~~bandwidth~~contraint\\
~~cloud~~computing~~contraint\\
~~budget~~contrant
$$

We can formulate as a cooperative game

## Solutions
- Problem is NP-Hard
- Build 1-2 algorithms to solve
- Need some theorems
- Target submission venue: TON.
## TODO
- **7/6/2020**:
- @dinhquangthinh: arrange reference list in descending order of importance.
- @dinhquangthinh & Hưng Ngô: literature review + initial problem formulation.
- @Anh Hưng:
- need some theorectical stuffs in this papers, need your advise on performance guarantees, or proof of convergence.
- Need your advise whether we should choose a loss function to minimize loss or to maximize a score function for the convenience of making some theories.
- Black box solutions are not prefered in IEEE Networking people
- relation between features and computing load, could be linear?
- **TODO**: Discuss with Linh Nguyễn (Inverse optimization problems e.g. Selective Sensing), Viên Ngô & Hùng Lê (submomdularity).
- @Anh Phương:
- Need some references about feasible applications
- Already built on applications where satellite data and other data (sensors, UAV) are combined to build an application

## Progress Updating

1. Litterature Review
2. Define the problem formulation
3. Learn techinques to solve the problem
4. Simulate to examine techniques

## Future Work
Distributed computing in hyperspectral image proscess

có thể dùng nhiều uav cũng 1 ability nhưng mà collect các phổ khác nhau

sieu pho can tốc độ truyền cao

push broom

train ML tại chỗ

Flux Tower, đo carbon change, eco, Database NEON
Phenocam

Link dữ liệu NEON với vệ tinh, vì vệ tinh thì rộng về diện tích, capture được spatial varation, còn các trạm mặt đất thì nó không capture được spatial variation, cho nên thông thường sẽ dựa vào thông tin sensor từ trạm mặt đất ở vài điểm rồi dựa vào corelation với spectral data, mình generalize cho toàn bộ khu vực. Tương tự, với ảnh vệ tinh có ảnh thì chụp liên tục nhưng resoluation thấp, có ảnh thì chụp ít frequent hơn nhưng resolution cao, kết hợp cả 2 để vừa resolution cao vừa recency

Khi mà deloy ml trên flux tower thì ai được quyền access vào cái inference results, detect object,

nếu broadcast

có tháp của tư nhân không? thì có thể có [federated learning](https://paperswithcode.com/search?q_meta&q=federated%20learning&fbclid=IwAR1vZx9_7o-ou-DNugVMph2sCylM30t8mc9eHlIViQmiruLPFktfrONzx-U)

Xem xem có thể formulate thành 1 bài submodular

Assume that there are several organizer need to cooperate together, each org has its own data and infrastructure to process data

sensor hyperspectral: ghi du lieu 2 nm, cu 1 band 2nm la lại ghi. Do phan giai pho.

anh da pho thi nhan dc nang luong lon nhưng độ sensitivity lại giảm.

3 questions need to answer?
- How much data each org need to collect, store and process in federated environment
- with that amount of data, how much communication, computing and storage cost each org need to pay
- How federated learning have org to reduce cost, while keeping some performance, such as accuracy
- If orgs'data are different, how can we leverage that
- If orgs' data are similar, how can we leverage that
## Main references
- [Toward 6G Networks: Use Cases and Technologies](https://ieeexplore.ieee.org/document/9040264), Marco Giordani, Michele Polese, Marco Mezzavilla, Sundeep Rangan, and Michele Zorzi. IEEE Communications Magazine 2020
- [Space-Air-Ground IoT Network and Related Key Technologies](https://ieeexplore.ieee.org/document/8961915), Tao Hong, Weiting Zhao, Rongke Liu, and Michel Kadoch. IEEE Wireless Communications 2020
- [Airplane-Aided Integrated Networking for 6G Wireless](https://ieeexplore.ieee.org/document/8760401), Xiaojing Huang, J. Andrew Zhang, Ren Ping Liu, Y. Jay Guo, and Lajos Hanzo. IEEE Vehicular Technology Magazine 2019
- [Satellite-Terrestrial Integrated Edge Computing Networks: Architecture, Challenges, and Open Issues](https://ieeexplore.ieee.org/document/9048610), Renchao Xie ; Qinqin Tang ; Qiuning Wang ; Xu Liu ; F. Richard Yu ; Tao Huang. IEEE Network 2020
- [Energy Efficient Data Collection and Device Positioning in UAV-Assisted IoT](https://ieeexplore.ieee.org/document/8894454), Zijie Wang, Rongke Liu, Qirui Liu, John S. Thompson, Michel Kadoch. IEEE Internet of Things Journal 2020
- [Fusion of Spectral and Spatial Information for Classification of Hyperspectral Remote-Sensed Imagery by Local Graph](https://ieeexplore.ieee.org/document/7342891),Wenzhi Liao, Mauro Dalla Mura, Jocelyn Chanussot, Aleksandra Pižurica. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 2016
- [Feature Extraction for Hyperspectral Imagery: The Evolution from Shallow to Deep (Overview and Toolbox)](https://ieeexplore.ieee.org/document/9082155). Behnood Rasti, Danfeng Hong, Renlong Hang, Pedram Ghamisi, Xudong Kang, Jocelyn Chanussot, Jon Atli Benediktsson. IEEE Geoscience and Remote Sensing Magazine 2020
- [Fast Parallel Algorithms for Statistical Subset Selection Problems](https://papers.nips.cc/paper/8751-fast-parallel-algorithms-for-statistical-subset-selection-problems). Sharon Qian, Yaron Singer. NIPS'19.
- [Service Placement and Request Routing in MEC Networks With Storage, Computation, and Communication Constraints](https://ieeexplore.ieee.org/document/9055440), Konstantinos Poularakis ; Jaime Llorca ; Antonia M. Tulino ; Ian Taylor ; Leandros Tassiulas. TON2020. Simulation code available (quite simple).

https://www.researchgate.net/publication/340071152_Latency_and_Energy_Optimization_for_MEC_Enhanced_SAT-IoT_Networks

https://en.wikipedia.org/wiki/CubeSat

Service Placement and Request Routing in MEC
Networks With Storage, Computation, and
Communication Constraints, ACM/IEEE TON

AUC Maximizing Support Vector Machines with Feature Selection

Online AUC Maximization

R. Iyer and J. Bilmes. "Algorithms for approximate minimization of the difference between submodular
functions, with applications". In UAI, 2012

Iyer, R. K., and Bilmes, J. A. 2013. "Submodular optimization with submodular cover and submodular knapsack constraints". In Advances in Neural Information Processing Systems, 2436–2444.

Mike Robert et. al. "Submodular Trajectory Optimization for Aerial 3D Scanning", ICCV 2017

S. Jegelka and J. Bilmes. "Online submodular minimization for combinatorial structures". ICML, 2011

A. Krause and C. Guestrin. "Near-optimal nonmyopic value of information in graphical models." In
. UAI, 2005

Submodularity in Data Subset Selection and Active Learning

iCrowd: Near-Optimal Task Allocation
for Piggyback Crowdsensing

Algorithms for Approximate Minimization of the Difference
Between Submodular Functions, with Applications

On Subset Selection with General Cost Constraints, IJCAI

Fast Parallel Algorithms for Statistical Subset
Selection Problems, NIPS 2019