Feature Selection for Intelligent Firefighting Robot Classification of Fire, Smoke, and Thermal Reflections Using Thermal Infrared Images

Table 1

Conventional and vision-based features.

Type	Feature	Advantages	Disadvantages
Conventional features [6, 7]	Temperature; Ionization; UV light	(i) Detect presence of fire and smoke [8]	(i) Long response time [8]; (ii) Unable to provide sufficient data for fire locating
Model-based; features	Fourier transform [20]; Wavelet transform [9]	(i) Frequency content analysis; (ii) Flexible analysis of both space and frequency [25]	(i) Unable to be spatially localized [25]
Vision-based features	Color (RGB) [9–12, 26]	(i) Fire (red); (ii) Smoke (gray)	(i) RGB camera cannot function in smoke-filled environments [2, 14]
	Dynamics [13, 14] (motion, shape change, etc.)	(i) Flickering flames recognition; (ii) Smoke flow detection	(i) Can be influenced by dynamical robot motion; (ii) Expensive computation for motion compensation
	Texture [12, 18, 19, 27]	(i) Spatial characteristics for pattern recognition; (ii) Less influenced by rotation and motion [18]	(i) The higher the order texture features, the more the computation
	Feature maps [28] (CNN deep learning)	(i) Superior performance in pattern recognition [29]; (ii) Once trained, applicable in real-time	(i) Slow learning speed; (ii) GPUs required due to expensive computation