Deep Learning Approach for Building Detection Using LiDAR–Orthophoto Fusion

Table 1

Extracted features from the LiDAR and orthophoto data [16–19].

Data source	Feature group	Feature	Description
Orthophoto	Spectral	Mean red	Average value of the pixels that cover the segment in the red band
		Mean green	Average value of the pixels that cover the segment in the green band
		Mean blue	Average value of the pixels that cover the segment in the blue band
	Texture	GLCM angular
		GLCM contrast
		GLCM correlation
		GLCM dissimilarity
		GLCM entropy
		GLCM homogeneity
		GLCM mean
		GLCM variance
LiDAR	Shape	Area	Total area of segment without holes
		Compactness	Ratio of the area of a polygon to the area of a circle with the same perimeter
		Density	Distribution in space of the pixels of an image object
		Length/width	Length-width ratio of the envelope rectangle
		Rectangular fit	Goodness of a building that fits into a rectangle
		Roundness	Area of the segment to the square of the maximum diameter of the referred segment
		Shape index	Border length of the segment divided by four times the square root of its area
	LiDAR	DEM	Digital elevation model
		DSM	Digital surface model
		nDSM	Object height by subtracting DEM from DSM

In the equations above, is the row number of the cooccurrence matrix, is the column number of the cooccurrence matrix, and is the normalized value in cell (), where is the value in cell of the cooccurrence matrix and is the number of rows or columns of the cooccurrence matrix.