2.3.4.2  Remote sensing image classification
 
 
Remote sensing image interpretation is the process of analysis, reasoning and judgment based on feature  information of various identify targets provided by remote sensing images, to identify the target surface features or phenomena, so as to obtain  information target surface features. This study extracts remote sensing image information by combining computer initial classification and visual interpretation.The computer automatically classification is completed by combining ERDAS and eCognition. Based on classification techniques of spectral information facing pixel, ERDAS can extract vegetation information small surface features very well, and conduct image processing with unsupervised method (examples shown in Figure 2-6). Multi-scale image segmentation technology of eCognition makes up for defects on classification scale of remote sensing classification software, which is based on  traditional spectral information facing pixel.
 

2.3.4.3  Visual interpretation
 
 
Modify initial classification results of remote sensing, and perfect land that are wrongly classified or left out. Specific operation is to identify target lands according to interpretation  signs  of images (Table 2-1) , such as hue (color), shape, location, size, shadow, texture, and other indirect signs and personal familiarity with the distribution of land in the region. At the same time, draw changed land type along the edge of image feature accurately (boundary should be strictly closed). The minimum unit of target surface features interpretation and extraction: according to general requirements,  variable land should be larger than 3 × 3 pixels, the minimum width of shorter edge of pattern spot two pixels. The interpretation accuracy of pattern spot elements should be > 85.0%, visual interpretation line tracing accuracy should be two pixel points and keep mellow. Diagram shown in Figure2-7

Correction of TM data interpretation results

 
Figure  2-7 Picture of visual interpretation
 
2.3.4.5  Accuracy test
 
 
Conduct accuracy test on remote sensing classification results by fixed sample location data of continuous forest inventory in 1979 and 1992. The distribution of MSS, TM testing points are shown in Figure 2-8 and Figure 2-9. Accuracy of remote sensing classification is tested through the establishment of error matrix of land use types.
 
It can be seen from Table 2-2 and 2-3 that the classification accuracy of woodland and farmland is higher, while that of construction land and unused land is lower, because the minimum pattern spot of interpretation results is set to 90 * 90m2, resulting in information lost of construction land, the area of which is smaller than the set value; the reason that classification accuracy of unused land is low is because the definitions of unused land are different in two classification systems.
 
Table  2-2                        Error matrix of MSS classification                   
 

			Remote sensing classification results
			Woodland	Farmland		Grass land	Water area		Unused land	Construction land		Total
Sample plots data	Woodland		1184	278		160	1		18	1		1642
	Farmland		368	1188		13	21		1	5		1596
	Grass land		365	81		576	2		105			1129
	Water area		14	23		4	23		3			67
	Unused land		76	13		86			49			224
	Construction land		27	47		7	3			4		88
	Total		2034	1630		846	50		176	10		4746
Overall accuracy: 63.72%
		Mapping Accuracy			Omission Accuracy			User Accuracy			Misclassification error
Woodland		72.11%			27.89%			58.21%			41.79%
Farmland		74.44%			25.56%			17.06%			82.94%
Grass land		51.02%			48.98%			18.91%			81.09%
Water area		34.33%			65.67%			2.00%			98.00%
Unused land		21.88%			78.13%			10.23%			89.77%
Construction land		4.55%			95.45%			1--%			9--%
Kappa:47.56%（N=4746）

 
Table  2-3                                          Error matrix of TM classification
 

		Remote sensing classification results
		Woodland	Farmland	Grass land	Water area	Unused land	Construction land		Total
Sample plots data	Woodland	2893	687	357	9	25	8		3979
	Farmland	550	1730	15	20	2	34		2351
	Grass land	183	8	533	8	20	1		753
	Water area	16	45	3	51	2	2		119
	Unused land	217	16	474	6	148	2		863
	Construction land	22	71	0	7	0	14		114
	Total	3881	2557	1382	101	197	61		8179
Overall accuracy: 65.64%
Woodland		72.71%		27.29%		74.54%		25.46%
Farmland		73.59%		26.41%		67.66%		32.34%
Grass land		70.78%		29.22%		38.57%		61.43%
Water area		42.86%		57.14%		50.50%		49.50%
Unused land		17.15%		82.85%		75.13%		24.87%
Construction land		12.28%		87.72%		22.95%		77.05%

 
 
2.4 Other statistics and survey data
 
 
Including statistical data of afforestation, forest management and other forestry activities, and statistical data of afforestation, harvesting, disease and pest, fire, etc. in forestry statistical yearbook (1978-2012), as well as statistical data of protection forest project in the upper reaches of the Yangtze River, natural forest protection project, reforestation project etc.
 
2.5 Land use transfer matrix
 
 
Land use transfer matrix obtained from the above data. For methodology of the first and second level, the matrix is land type transfer matrix during each inventory, for methodology of the third level, annual transfer matrix can be obtained. As an example, Table 2-4 shows land type transfer matrix during the 2007-2012 inventory.

Unitrans世聯(lián)翻譯公司在您身邊,離您近的翻譯公司,心貼心的專業(yè)服務(wù),專業(yè)的全球語(yǔ)言翻譯與信息解決方案供應(yīng)商,專業(yè)翻譯機(jī)構(gòu)品牌。無(wú)論在本地,國(guó)內(nèi)還是海外,我們的專業(yè)、星級(jí)體貼服務(wù)，為您的事業(yè)加速！世聯(lián)翻譯公司在北京、上海、深圳等國(guó)際交往城市設(shè)有翻譯基地，業(yè)務(wù)覆蓋全國(guó)城市。每天有近百萬(wàn)字節(jié)的信息和貿(mào)易通過(guò)世聯(lián)走向全球！積累了大量政商用戶數(shù)據(jù)，翻譯人才庫(kù)數(shù)據(jù)，多語(yǔ)種語(yǔ)料庫(kù)大數(shù)據(jù)。世聯(lián)品牌和服務(wù)品質(zhì)已得到政務(wù)防務(wù)和國(guó)際組織、跨國(guó)公司和大中型企業(yè)等近萬(wàn)用戶的認(rèn)可。 專業(yè)翻譯公司,北京翻譯公司,上海翻譯公司,英文翻譯,日文翻譯,韓語(yǔ)翻譯,翻譯公司排行榜,翻譯公司收費(fèi)價(jià)格表,翻譯公司收費(fèi)標(biāo)準(zhǔn),翻譯公司北京,翻譯公司上海。