FireXnet: an explainable AI-based tailored deep learning model for wildfire detection on resource-constrained devices

Fire Ecology

Table 8 Confidence intervals and Cohen’s kappa coefficient for all models

Models	True labels	Predicted labels				Accuracy (%)	95% CI	Kappa
		Fire	No Fire	Smoke	Thermal
The proposed FireXnet model	Fire	93	0	2	0	98.42	[0.97, 1.00]	0.98
	No Fire	1	92	2	0
	Smoke	0	1	94	0
	Thermal	0	0	0	95
DenseNet201	Fire	90	0	5	0	97.63	[0.96, 0.99]	0.97
	No Fire	1	93	1	0
	Smoke	2	0	93	0
	Thermal	0	0	0	95
VGG16	Fire	90	3	2	0	94.75	[0.92, 0.96]	0.95
	No Fire	2	89	4	0
	Smoke	3	6	86	0
	Thermal	0	0	0	95
MobilenetV2	Fire	91	0	4	0	98.16	[0.97, 1.00]	0.98
	No Fire	1	92	2	0
	Smoke	0	0	95	0
	Thermal	0	0	0	95
Inception-ResNetV2	Fire	92	1	2	0	96.05	[0.95, 0.98]	0.96
	No Fire	2	92	1	0
	Smoke	4	5	86	0
	Thermal	0	0	0	95
InceptionV3	Fire	92	1	2	0	96.84	[0.95, 0.98]	0.97
	No Fire	1	88	6	0
	Smoke	2	0	93	0
	Thermal	0	0	0	95