Data Fusion - Science topic

The fatigue tests are conducted under tensile-tensile loading with R = 0.1, and electrical resistance was measured in real time during the fatigue test. We recorded force, displacement, and electrical resistance at predefined cycle intervals. The goal is to understand how mechanical energy dissipation (hysteresis loop area) correlates with electrical resistance evolution over fatigue life, particularly under different applied voltages.

I sent an email to the professor M. Bilal, until today I have not received a reply. The draft of the student program are interested and I would like to organize something similar at our faculty ECPD in Belgrade and to achieve cooperation with King Fahd Universuty.

Regards

PhD Tomislav Đorđević arch

Dear,

Anatol Badach,

Thank you for sharing these valuable resources on Digital Twins (DTs) in construction. I'll explore them to deepen my understanding and integrate their insights into my research on IoT and Big Data Fusion in Construction MIS. DTs hold great promise for revolutionizing construction practices, enhancing efficiency, safety, and sustainability.

Best regards,

Mithun Kokare.

In the context of multi-static radar systems, data fusion is a critical process that involves integrating information from multiple sensors to enhance target detection, tracking, and environmental perception. The necessity of using data from the same transmitter for fusion in such systems is not explicitly required but can depend on the specific application and system design. The integration of data from different transmitters can be beneficial in certain scenarios, as highlighted by various research findings:

1. **Cooperative Fusion for Enhanced Detection**: A study on passive multistatic radar systems highlighted the use of cooperative fusion techniques, including both hard and soft fusion, to improve target detection. These techniques do not necessarily require data from the same transmitter but leverage spatial diversity across multiple receivers. The study proposed novel fusion techniques that showed significant detection performance improvements without the need for knowledge of the transmitted signal or channel information [1]

2. Benefit Analysis of Data Fusion: Another research conducted a benefit analysis of data fusion for target tracking in multiple radar systems (MRS). It suggests that data fusion, whether from the same or different transmitters, can enhance tracking performance depending on factors like signal-to-noise ratio, deployment, and resolution of each radar. This implies that the necessity of fusing data from the same transmitter is not a strict requirement but should be considered based on the potential performance enhancement [2]

3. Linear Fusion Framework: A linear fusion framework for target detection in passive multistatic radar systems was proposed to improve detection performance. This framework involves a weighted combination of local test statistics from spatially separated receivers, suggesting that integrating data from different transmitters can be advantageous [3]

4. Multifrequency GPR Data Fusion: In the context of ground-penetrating radar (GPR), a novel method was developed for the fusion of data from antennas operating at different frequency ranges. This approach, focusing on enhancing subsurface imaging, illustrates the benefit of combining data from different sources, analogous to fusing data from different transmitters in a radar context [4]

5. Multi-Target Tracking in Passive Systems: A study on multi-target tracking in passive multi-static radar systems using Doppler-only measurements discusses the advantages of fusing measurements from spatially distributed sensors. It highlights the fusion of data from multiple bistatic links, again indicating that the fusion process can benefit from integrating information from different transmitters [5]

6. Hybrid Radar Fusion: [6] introduces "hybrid radar fusion" within an integrated sensing and communication scenario. It involves a dual-functional radar and communications base station performing as a mono-static radar for sensing in the downlink, while also handling communication tasks. Communication users act as bi-static radar nodes in the uplink. The study focuses on fusing information from different resource bands to estimate angles-of-arrival for multiple targets, proposing efficient algorithms for this purpose and demonstrating their performance through simulations.

References

[1] Asif, Asma, and Sithamparanathan Kandeepan. "Cooperative fusion based passive multistatic radar detection." Sensors 21.9 (2021): 3209.

[2] J. Yan, H. Liu, W. Pu, B. Jiu, Z. Liu and Z. Bao, "Benefit Analysis of Data Fusion for Target Tracking in Multiple Radar System," in IEEE Sensors Journal, vol. 16, no. 16, pp. 6359-6366, Aug.15, 2016, doi: 10.1109/JSEN.2016.2581824.

keywords: {Radar tracking;Target tracking;Data integration;Signal to noise ratio;Bayes methods;Radar measurements;Multiple radar system;target tracking;Bayesian Cramér-Rao lower bound;benefit analysis},

[3] Zhao, Hong-Yan, et al. "Linear fusion for target detection in passive multistatic radar." signal Processing 130 (2017): 175-182.

[4] A. De Coster and S. Lambot, "Fusion of Multifrequency GPR Data Freed From Antenna Effects," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 2, pp. 664-674, Feb. 2018, doi: 10.1109/JSTARS.2018.2790419.

keywords: {Ground penetrating radar;Radar antennas;Data integration;Antenna measurements;Mathematical model;Antenna effect removal;full-wave inversion;Green's functions;ground-penetrating radar (GPR) multifrequency antenna data fusion},

[5] S. Subedi, Y. D. Zhang, M. G. Amin and B. Himed, "Group Sparsity Based Multi-Target Tracking in Passive Multi-Static Radar Systems Using Doppler-Only Measurements," in IEEE Transactions on Signal Processing, vol. 64, no. 14, pp. 3619-3634, 15 July15, 2016, doi: 10.1109/TSP.2016.2552498.

[6] A. Chowdary, A. Bazzi and M. Chafii, "On Hybrid Radar Fusion for Integrated Sensing and Communication," in IEEE Transactions on Wireless Communications, doi: 10.1109/TWC.2024.3357573.

keywords: {Radar;Sensors;OFDM;Robot sensing systems;Wireless communication;Fuses;Wireless sensor networks;Integrated Sensing and Communication (ISAC);Dual-Functional Radar and Communications (DFRC);radar fusion;hybrid radar;6G},

.

Useful thing

Received, thank you for your answer。

Dear Dr. Dejan

Thank you very much for the information and for asking for research collaboration. If you wish to work with me, please send your abstract first. I am a member of the review board in Land, MDPI. I will check the possibilities for a discount or free publication. Thanks

Dear Saif M. Abo Khashaba .

The integration of data and knowledge from several sources is known as data fusion. This paper summarizes the state of the data fusion field and describes the most relevant studies. We first enumerate and explain different classification schemes for data fusion. Then, the most common algorithms are reviewed. These methods and algorithms are presented using three different categories: (i) data association, (ii) state estimation, and (iii) decision fusion.

Regards,

Shafagat

Sensor data fusion is used to take advantage of multiple sensors, generally based on different modality, to accomplish a goal better than any individual sensor would. For example, SLAM algorithm usually rely on sensor data fusion from camera, IMU and sometimes other inputs (lidar, ultrasound, etc.) in order to better localize and model the environment around an vehicle (or UAV, etc.)

The term sensor stitching is more used to extend the spatial range of a single modality with multiple sensor. The best example is using multiple photos with overlapping areas to create a bigger panoramic image.

I hope these two examples help you better understand the distinction between these related concepts.

Application example relying on numerically generated Gaussian white noise ― While investigating adaptive control and energetic optimization of aerobic fermenters, I have applied the recursive least squares algorithm (RLS) with forgetting factor (RLS-FF) to estimate the parameters from the K_La correlation, used to predict the O₂ gas-liquid mass-transfer, while giving increased weight to most recent data. Estimates were improved by imposing sinusoidal disturbance to air flow and agitation speed (manipulated variables). The proposed (adaptive) control algorithm compared favourably with PID. Simulations assessed the effect of numerically generated white Gaussian noise (2-sigma truncated) and of first order delay. This investigation was reported at (MSc Thesis):

Visit to read this article

Thank you, @Sarvat Gull

You can use SPSS manual and go to to " Transform" and under transform click "recode to different variables". The SPSS MANUAL CAN BE FOUND ONLINE.

Thank you so much, Erick. I will contact them.

Hi Debakanta,

You have to register in the IEEE GRS contest website to get the dataset. Check this URL:

Good Luck

Dear Mahdis Dezfouli,

Data fusion is the process of getting data from multiple sources in order to build more sophisticated models and understand more about a project. It often means getting combined data on a single subject and combining it for central analysis.

or

Data fusion frequently involves “fusing” data at different abstraction levels and differing levels of uncertainty to support a more narrow set of application workloads.

or

Data fusion is the process of integrating multiple data sources to produce more consistent, accurate, and useful information than that provided by any individual data source.

Various types of data fusion work in different ways:

Low, intermediate and high-level data fusion – and likewise distinguish geospatial types of data fusion from other types of data fusion. Another specific type of data fusion is called “sensor fusion” where data from diverse sensors are combined into one data-rich image or analysis.

Data fusion is broadly applied to technologies, for instance, in a research project, scientists might use data fusion to combine physical tracking data with environmental data, or in a customer dashboard, marketers might combine client identifier data with purchase history and other data collected at brick-and-mortar store locations to build a better profile.

Data fusion involves a level of concrete definition from something called the Joint Directors of Laboratories Data Fusion Group which produces six levels for a data fusion information group model:

02. Data integration :

Data integration is a process in which heterogeneous data is retrieved and combined as an incorporated form and structure. Data integration allows different data types (such as data sets, documents and tables) to be merged by users, organizations and applications, for use as personal or business processes and/or functions.

or

Data integration is the combination of technical and business processes used to combine data from disparate sources into meaningful and valuable information. A complete data integration solution delivers trusted data from various sources.

or

Data integration involves combining data from several disparate sources, which are stored using various technologies and provide a unified view of the data. Data integration becomes increasingly important in cases of merging systems of two companies or consolidating applications within one company to provide a unified view of the company's data assets. The later initiative is often called a data warehouse.

or

Data integration in the purest sense is about carefully and methodically blending data from different sources, making it more useful and valuable than it was before. IBM provides a strong definition, stating “Data integration is the combination of technical and business processes used to combine data from disparate sources into meaningful and valuable information.”

An example of data integration in a smaller paradigm is spreadsheet integration in a Microsoft Word document.

I hope I have answered your question.

With Best Wishes,

Samir G. Pandya.

i suggest having a look at the book:

'Mathematics of Data Fusion' by Goodman, Mahler and Nguyen

if i am not mistaken the theory of random finite sets formalised by mahler subsumes Dempster-Shafer theory, so it is worth having a looking into.

I do not understand your question well. What is your observation?

You can also check this paper; it was part of my thesis.

Experiments with Simultaneous Environment Mapping

and Multi-Target Tracking

Dear Dr.

I need to fuse heterogeneous data for different applications but first one is an environmental application which uses mainly continuos and discrete variables. The main task is the prediction of pollutant variables.

thanks in advance

M.A. Becerra

The human sensory system: view, hearing, smell, taste, touch, is a paradigm for data fusion. In strict sense, for fusion you need at least two sensors for essentially different signals,

About the Kalman filter number of sensors needed, you should first have the models (system and observations models), taking into account their Controllabilty and Observability; This reference tackles this point:

Controllability and Observability: Tools for Kalman Filter Design;

B Southallz y , B F Buxtony and J A Marchantz

London, London WC1E 6BT, UK

Take a look at the text and papers given in the links below; perhaps they might

be of interest to you (it is an answer I gave to a previous questions):

Attached are links to papers I published related to this topic that may be of interest to you (they are relatively old papers on work I did 2 to 3 decades ago).  It will show you an example of kind of 'pushing' the limits of merging very different spatial resolution image data sets.  Keep in mind that usually the spectral information if what the lower resolution data set contributes and spatial resolution (or detail) is what the higher resolution image data set contributes.  Also, I have a paper that discusses how merging multi-resolution image data sets together do better is they cover the similar spectral range to minimize edge reversals.   1991 Paper Comparison of Three Different Methods to Merge Multiresolution and Multispectral Data: Landsat TM and SPOT Panchromatic https://www.researchgate.net/publication/200458993_Comparison_of_Three_Different_Methods_to_Merge_Multiresolution_and_Multispectral_Data_Landsat_TM_and_SPOT_Panchromatic 1986 Paper Digital merging of Landsat TM and digitized NHAP data for 1: 24,000-scale image mapping((National High Altitude Program)) https://www.researchgate.net/publication/240321338_Digital_merging_of_Landsat_TM_and_digitized_NHAP_data_for_1_24000-scale_image_mappingNational_High_Altitude_Program Pat Article Comparison of Three Different Methods to Merge Multiresoluti... Article Digital merging of Landsat TM and digitized NHAP data for 1:...

Hello,

the problem you are referring to usually receives the name of "inertial navigation" or "dead reckoning". It is something I am currently working on, and is a problem whose solution is not direct.

The method I am using is:

1. Obtain a good "orientation estimation" or "attitude estimation". This is, find the best guess of the rotation transformation that relates two reference frames: the one attached to your sensor, and other one usually taken to be an inertial reference frame (or approximately inertial) as one whose z-axis points in a direction orthogonal with the Earth surface. The accelerometers sense the gravity accelerations, so it gives you information about "where it is down". Gyroscopes help predict orientation. A common approach is to use a Kalman filter:

although others use gradient descent techniques and also obtain good results:

2. Transform the acceleration measurements (taken in a non inertial reference frame: the one attached to your sensor) to the inertial reference frame using the rotation transformation obtained with orientation estimation.

3. Subtract the acceleration due to gravity.

4. Integrate the resulting acceleration twice to obtain position.

This is what you usually obtain when you perform this process:

If you have seen those videos you will have begun to think that it is not possible. However, in certain scenarios you can use pseudo-measurements to improve your estimation:

I would conclude that depending on your application, you should use pseudo-measurements, or include more sensors to estimate the position. In other case, as far as I know, it is not possible to obtain a good position estimation.

Ok thank you so much Mr Ali Farmani

Hi Ariel,

As Graham mentioned, that you will need to add process noise in order for the filter to converge. Also assuming H as identity can lead to bad results as you always assume that the GNSS and IMU are bias-free (which is real life is not the case). I would suggest adding the bias in the GNSS and IMU data as state variables for the filter to estimate them as well. 

You can have a look at our work regarding the same. If you want I can also attach you c++ implementation of the same.

Regards.

What type of hysteresis sensor are you studying?

Magnetic?

Refer to the attached discussions. It might be helpful.

Thanks Enrique. Well, your assumptions are true but weights are connected to and calculated based on the measurements vector. Let's say that we have the state vector you proposed above with a measurement vector has ranges and rate of change of ranges as measurements which are associated with the position and speeds in the sates vector. Now, we you calculate the weights using a measure of the distance between the measured quantities and the predicted ones. Here you find that you are getting numbers in very deferent ranges (order of magnitude) since the measured quantities are in different domains i.e. range and rate of change of range. When you normalize these calculated weights, the smaller ones will vanished and have no effect the process. So, the question again, or what I found is that I need to calculate the weights separately, normalize them, and then merge them together but wanted someone to confirm this if they noticed same thing.  

@Mohamed Thanks a lot for your suggestion.

Thank you for everyone's help! I am now studying the papers.

Your are quite welcome, Sir.

Thanks for the answer!

Didn't find in this arcticle multisensor fusion without using cross-covariances though. Not sure it is possible to combine techniques from this article and "Zhen Ding, Lang Hong " A distributed IMM fusion algorithm for multi-platform tracking " (1998)" to deal with asynchronous IMM.

Dear Mohsen,

I suggest to you links and attached files in topics.

-Robust car tracking using Kalman filtering and Bayesian templates ...

proceedings.spiedigitallibrary.org/proceeding.aspx?articleid...

-How a Kalman filter works, in pictures | Bzarg

-computer vision - Intuitive explanation of tracking with Kalman filters ...

dsp.stackexchange.com/.../intuitive-explanation-of-tracking-with-k...

Best regards

Well Aparna,

This is becoming too vague for me. Convince me with a Matlab program that works for the two (I admit, very different) examples uploaded by Lucas.

Harris, cartesian fractals, I don't care. Just segment the images.

Hello. Manoeuvring target tacking is a difficult area. There are really a lot of approaches. If you listen to Bar-Shalom et al, you will be using interacting multiple model, which itself is just an approximation to a Bayesian filer for a hybrid dynamical system. Singer is a more conventional approach. I have also used RHV coordinates (in Blackman's book). The choice of coordinates is really important. The filter is only as good as the dynamical model it's based on. For ATC you should be able to get a pretty good dynamical model (centred on the axes of the aircraft) and if you have 3-D measurements, you should be able to set the problem up pretty well. You will need good measurement gating (which should occur naturally if you have body centred coordinates).

Here are some references:

For the record, the EM (Viterbi+Kalman smoother) approach is very high performance (better than IMM) but does not explicitly allow for clutter (except through a PMHT-like approximation), but should be good for low clutter or in higher dimensional measurement space (read dim > 1).

The survey paper, which would have been the first comprehensive paper on this topic, was supposed to appear in the IEEE Transactions on Aerospace & Electronic Systems in 1998, but was suppressed by the associate editor at the time, whose student was working on ... manoeuvring target tracking! My paper (based on work done in 1996) was in review for 3 years and by the time I got the reviews back it was obsolete! I did complain to the E-I-C of IEEE-TAES but you can't go backwards in time. Now you know why I am not recommending certain work by a person whose name should be obvious to you!

Good luck!

Please go through chapter -2 (especially page number 96) of text book: Multitarget- Multisensor Tracking: Principle and Techniques (1995) by Barshalom and X. Rong-li

Finally i found that sentine-1 could be combine with Radiometer data like SMOS.

I think there is no way to fused sentinel with optic data right now.

you could take a look to our simulator here

Good luck!

first see the wavemenu in matlab. it is a simple GUI (Graphical User Interface). you can understand how by using wavelets you can fuse two signals. 

find the 1-D DWT for two signals separately and then add their coefficients. finally find the 1-D IDWT for the coefficients. 

by this you can develop the dataset. 

i hope it is helpful to you.

making sensors sends only abnormal readings only for (reading rand: max and min values). each gateway saves range of readings for each sensor and it informs the root gateway if there is an effective change (for example implementing 50th percentile on the historical data).

Do you mean that measurements may arrive out of sequence? So that newer data may arrive at your tracker before older data?

Rather than trying to wrangle with the maths (using prediction and 'retro-diction') could you just push all of your measurements through a buffer? Sort them as they pass through. Then apply the IMM/PDA algorithm on the sorted measurements at the other end? This would apply a lag/latency to your system (determined by the maximum expected transmission delay in your network), but if that is tolerable it might be a simple way of dealing with your problem.

dear carlos

this is a link for ieee paper just from 2015 American Control Conference

the abstract is great you can read it first:

Measurement censoring, or Tobit model censoring, is common in many engineering applications. It arises from limits in sensor dynamic range, and may be exacerbated by poor calibration of sensors. Censoring is often referred to as a clipped measurement or limit-of-detection discontinuity, and is represented as a piecewise-linear transform of the output variable. The slope of the piecewise-linear transform is zero in the censored region. This form of nonlinearity presents significant challenges when a nonlinear approximation to the Kalman filter is to be used as an estimator. The Tobit Kalman filter is a new method that is a computationally efficient, unbiased estimator for linear dynamical systems with censored output. In this paper, we use Monte Carlo methods to compare the performance of the Tobit Kalman Filter to the performance of the Extended Kalman Filter and the Unscented Kalman Filter. We show that the Tobit Kalman Filter reliably provides accurate estimates of the state and state error covariance with censored measurement data, while both the EKF and the UKF provide unreliable estimates in censored data conditions

Best regards

Ashraf Abosekeen

Hi Peter,

After reading again your question I can confirm that this behavior is normal as you speak of error to the ground truth and not to uncertainty errors. (I was confusing myself between both is the previous messages).

As your state is a random variable depending on measurements, its meaning can be more or less accurate depending the observation.

Suppose that you have the ground truth x_gt = 2 and your estimate at time t is given by x_t = 2.1 with a variance of 4

At this step your mean estimate is very good with a poor variance. Suppose you have a measurement z = 3 with variance = 1.5 (the measurement is correct as the ground truth value is in its 3 sigma bound).

Then the fusion (update step) will move your current estimate to something higher than 2.1 making your distance to GT bigger but reducing the uncertainty.

At this step you cannot say that the result is poorer than the previous one because statistically it is much better.

So the behaviour you observed is not a bug, it's the normal behaviour of such filtering process as we speak of random variable. You have to consider both x and px.

Hoping my answer will help you.

If the measurements arriving asynchronously are independent, you can still use them to update the Kalman filter or EKF. Just predict to the required measurement time and filter using the appropriate H matrix for the measurement you have.

DF is improving DQ when (and only when) the different data input streams are to some degree correlated. If not, DF does not make any sense.

Sorry - I cannot give you academic materials because I have none on this topic. The above is from experience and some own work in this area.

Thanks Rasmus! Yes, I forgot to mention http://models.life.ku.dk/datasets, which is actually really useful if someone is looking for data (for multiblock analysis, multiway analysis or simply 2D)!

Another one for data fusion (QSAR) can be downloaded here: http://archive.ics.uci.edu/ml/datasets/QSAR+biodegradation

However, if someone knows other available sources, please share!

What do you mean by DTM ? Digital Terrain Map ? If so, a good way to generate these maps is the use of either parametric representations or the use of occupancy grids which can be 2,5D or even 3D, purely geometric and/or probabilistic. Interpolation methods to generate occupancy grid exist, some of them take into account lidar data uncertainties. Early in the 90's works of Takeo Kanade, In So Kweo, Martial Hebert, then Raja Chatila, Fawzi Nashashibi, Michel Devy, Simon Lacroix among others have published articles on that. Today's works in autonomous driving and autonomous mapping (e.g. CMU) are interesting. By the way what type of LIDAR are you using ? Single layer ? 4-layers ? Velodyne ?...

It is normal in a Kalman Filter to define and provide the error characteristics of each sensor. If you have reason to believe that the sensor errors vary with time or with some other, perhaps geometrical, property of the scenario, you should model that sensor behaviour. This is perfectly standard thing to do. If your model of the sensor behaviour is accurate, the standard KF will provide optimal results. 

We used eCognition too. It's also possible to implement your own calculations beside GLCM.

Hello. I think it depends on what you mean by unknown. Even the simplest estimators/filters make implicit assumptions about the errors in the system and measurement models, with most techniques requiring independent (white) noise. If you are after robust estimators, then the H-infinity class may be a good choice. I understand that this filter may not always be realisable depending on its tuning. Another class of estimator that does not make many assumptions about the data is set theoretic. In this case, you only need to bound the noise.

It doesnt work too!

I read their paper and send them emails and the result is same!

thank you

how does these sensors work,

their efficacy and limits,

what about sensor fusion???

@Laya Das Thank yo so much for your explnation. I am laso studying this issue and try to find the reason why the MHE is better. I'll update this question if I have something valuable.

Hi Cliff, a problem with making such a simulated dataset is that there not so much known on how you should introduce "noise" in a realistic fashion. Most researchers agree that the size of a cluster of duplicates is Zipf distributed and some have proposed some models for introducing typographical errors, but apart from that, it would be "guessing" what a realistic error model would be. For example, abbreviations, multi-valued attributes, subjectiveness (e.g. the musical genre of a CD)... And of course, if there is no realistic error model, simulating a dataset tends to be biased to work good on the algorithm you want to test.

If the range of the feature is different in different modalities and one is going to use some sort comparative analysis, it is always safe to normalise the features.

Moreover, often the feature set makes more sense and is easily interpretable in the normalized form.

Dear Marcelo,

please check the Wikipedia entry:

and some of the references at the end. Furthermore, I suggest to look at amazon.com for books on time-series analysis.

Good luck,

Reiner

Hello, I think it is really difficult to answer this question, because the navigation performance does not only depends on the data fusion approach but also the measurements you use. For example, I did the research on the autonomous navigation for spacecraft. The outputs from IMU, star sensor and navigation cameras are integrated by EKF to determine the r and v of the spacecraft. The results was very good. However, I was also involved in the demonstration of planetary landing navigation based on IMU, radio metric measurements, Doppler radar, and navigation camera. In this case, the EKF did not work well. I am also trying to solve this problem. But I believe that the EKF is suitable for most of issues.

Good luck! Thank you!

ns2 is better option because it is open source and wide range of tutorials are available if you are student then ns2 is best for you

I would strongly recommend a look at the work on decentralized (sensor) data fusion by Simon Julier and Jeffrey K. Uhlmann or for starters http://dsp-book.narod.ru/HMDF/2379ch12.pdf or even better the Handbook of Multisensor data fusion