**6. Conclusions**

Fire bursts are a dangerous problem of great importance worldwide. Mega fires often result in significant environmental destructions, major damages on infrastructures, and economic loss. Most importantly, they put at stake the lives, not only of the civilians but also of the forest fire personnel. Thus, technologies that facilitate early fire detection are important for reducing fires and their negative effects.

This chapter aims to provide an alternative view for early fire detection based on twitter posts, instead of expensive sensors and other infrastructures. A hybrid system architecture is introduced which combines a deep learning process for the detection of valid twitter posts regarding fire bursts and a NLP process which extracts the crucial information (place, time, etc.) from the valid tweets. Finally, risk assessment, based on analytics, is performed which derives the geographical places threatened by fire at the current time.

Part of the architecture is already validated under real-world conditions, and the results are promising. The overall system performance is expected to be further improved once the deep learning scheme is entirely utilized.

### **Acknowledgements**

This work was performed within the AF3 Project (Advanced Forest Fire Fighting), with the support of the European Commission by means of the Seventh Framework Programme (FP7), under Grant Agreement No. 607276.

#### **Conflict of interest**

There are no "conflict of interest" issues regarding this chapter.

#### **A. Appendices and nomenclature**

The mathematical definition of the convolution process between two onedimensional signals *f*(*t*) and *g*(*t*) follows in Eq. (9). The mathematics behind LSTM layer architecture follows in Eqs. (10)–(13). Functions *σ* and *tanh* represent the sigmoid and hyperbolic tangent function, respectively. Parameter W corresponds to weighting matrices:

$$(f\*\mathfrak{g})(t) = \int\_{-\infty}^{+\infty} f(\tau)\mathfrak{g}(t-\tau)d\tau\tag{9}$$

$$z\_t = \sigma(\mathcal{W}\_x \cdot [O\_{t-1}, O\_t])\tag{10}$$

**Author details**

Athens, Greece

**93**

Konstantinos-George Thanos\*, Andrianna Polydouri, Antonios Danelakis,

*Combined Deep Learning and Traditional NLP Approaches for Fire Burst Detection…*

*DOI: http://dx.doi.org/10.5772/intechopen.85075*

and Telecommunications, National Center for Scientific Research "Demokritos",

© 2019 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/ by/3.0), which permits unrestricted use, distribution, and reproduction in any medium,

Dimitris Kyriazanos and Stelios C.A. Thomopoulos

provided the original work is properly cited.

Integrated Systems Laboratory (ISL), Institute of Informatics

\*Address all correspondence to: giorgos.thanos@iit.demokritos.gr

$$r\_t = \sigma(\mathcal{W}\_r \cdot [O\_{t-1}, O\_t]) \tag{11}$$

$$h'\_t = \tanh(\mathcal{W} \cdot [r\_t \* O\_{t-1}, O\_t])\tag{12}$$

$$O\_t = (\mathbf{1} - z\_t) \* O\_{t-1} + z\_t \* h'\_t \tag{13}$$

*Combined Deep Learning and Traditional NLP Approaches for Fire Burst Detection… DOI: http://dx.doi.org/10.5772/intechopen.85075*
