An automatic extraction method for individual tree crowns based on self-adaptive mutual information and tile computing

Figures & data

Figure 1. Study areas and GeoEye-1 panchromatic images: (a) study area A is located east of Beihai City in Guangxi Province and (b) study area B is located east of Ningbo City in Zhejiang Province.

Figure 2. Flowchart for automatic extraction method based on SMI-TCT algorithm.

Figure 3. The organization of tile images.

Figure 4. The process for SMI algorithm.

Figure 5. Comparison of different values of new intensity level M; (a) M = 32; (b) M = 8, x and y axes are the number of tile images in different directions, z axis is MI value.

Table 1. The parameters of templates and MI thresholds in SMI algorithms.

Study area	Distribution of tree crowns	Templates dimensions (side length, m)	Number of templates	MI threshold
Beihai	Uniform	3–7.5	10	0.9
Ningbo	Nonuniform	3–8.5	20	0.85

Figure 6. Results of the tree crowns extraction for study area A: (a) attribute table includes the fields of tree crown diameters (diameter) and MI value (I_UV) and (b) zoomed image of typical eucalyptus forest.

Table 2. Accuracy of SMI-TCT and the four other algorithms.

		Beihai			Ningbo
Algorithms	Statistics	Regular	Mixed	Total	Regular	Mixed	Total
The SMI-TCT method	Correct	482	256	738	378	218	596
	Incorrect	29	38	67	53	61	114
	Missing	23	22	45	89	71	160
	Accuracy (%)	89.70	78.40	85.7	69.60	54.30	63.8
The fixed window method (Walsworth and King 1999)	Correct	455	226	681	344	199	543
	Incorrect	39	39	78	94	72	166
	Missing	50	52	102	123	90	213
	Accuracy (%)	82.30	67.40	77.0	53.60	43.80	49.9
The dynamic window method (Culvenor 2002)	Correct	463	248	711	393	207	600
	Incorrect	31	33	64	62	58	120
	Missing	42	30	72	74	82	156
	Accuracy (%)	85.6	77.4	82.6	70.8	51.7	63.5
The template matching method (Pollock 1996)	Correct	478	247	725	337	196	533
	Incorrect	23	32	55	86	75	161
	Missing	27	31	58	130	93	223
	Accuracy (%)	90.1	77.2	85.6	53.7	42.0	49.2
Multiscale tree crown extraction method (Brandtberg and Walter 1998)	Correct	434	242	676	353	210	563
	Incorrect	48	45	93	71	57	128
	Missing	71	36	107	114	79	193
	Accuracy (%)	76.6	70.7	74.5	60.4	52.9	57.5

Note: Number of reference tree crowns = correct + missing.