Automatic Detecting Rice Fields by Using Multitemporal Satellite Images,Land-parcel Data and Domain Knowledge

Printer Friendly Format

Page 3 of 3
| Previous |

Automatic Detecting Rice Fields by Using Multitemporal Satellite Images,Land-parcel Data and Domain Knowledge

Experiments
A traditional supervised classification using a single image was carried out first provide base of comparison. The test image is 193-6, which was collected when the rice was grown up. Using the maximum-likelihood classifier performed a pixel-based classification. Table 1 shoes the classification accuracy of the two tests. Although using the region-based classification provides higher accuracy than using the pixel-based classification, the both tests reveal the fact that it is inefficient to distinguish rice and non-rice fields by using a single image.

	User's acc.of rice	User's acc.of non-rice	Overall accuracy	K hat
Pixel based	72.7%	32.1%	62.3%	4.6%
Region based	69.7%	87.7%	74.4%	46.6%

Table 1 Accuracy assessment for single-image classification
Then, the proposed methods for detecting rice fields are tested by using the multimporal images of the same test site, and results are shown in table 2 the use images and detection criteria are listed as follows:

Profile Matching All of the test images were also used to perform matching. The threshold values of the cross-corrrlation function was0.9. The detection accuracy is shown in table 2
Peak detection All of the test images were also used. The threshold values op the peak height and width for detecting rice fields were 40 and 100(days). The detection accuracy is shown in table 2.
Difference classification There tests were completed for difference classification with respect to the uses of 2,3 and 4 images. The first test uses the difference of (NDVI_193-6 NDV_193-2);The second test uses the difference of (NDVI_193-6 NDV_193-2);and (NDVI_193-6 NDV_193-10); And the third test uses the difference of (NDVI_193-6 NDV_193-2); (NDVI_193-6 NDV_193-10); and (NDVI_193-6 NDV_193-11). The accuracy was improved when the number of the features of NDVI difference was increased.

	User's acc.of rice	User's acc.of non- rice	Overall accuracy	K hat
PM	95.3%	53.3%	84.1%	54.4%
PD	79.9%	83.0%	80.7%	70.3%
DC(2 images)	89.8%	70.7%	84.7%	60.7%
DC(3 images)	90.3%	75.3%	86.3%	65.2%
DC(4 images)	91.3%	76.8%	87.5%	67.9%

Table 2. Accuracy assessment for Multitemporal-image classification
The result of the region-based rice detection can be easily put into a geographical information system. A graphical presentation of the detection result and the correctness of detection can be shown as figure 7 by using a GIS tool.

Figure 7. A graphical presentation of the detection results of the DC method using 4 images.
Conclusions
A technology recognition of rice fields by using Multitemporal satellite images is proposed. The principal of this technology is applying a region-based classification by means of integrating geographical data and domain knowledge with Multitemporal Images.. this approach has several advantages in comparing to standard classification approaches:

The region-based classification directly generates inventory data related to land owners:
Referring to the domain knowledge, this approach will automatically identify rice fields by using multitemporal satellite images without the need of training data.
The use of multitemporal images will also increases then classification accuracy.
The region -based classification is more efficient in computing

According to the experiment tests, this approach has the potential to generate a comparable accuracy to the traditional photogrammetric approach. Comparing to the traditional supervised classification using a single image epoch, all the methods can easily improve the accuracy about 20%. The PM and PD methods can also determine the planting time of a rice field, but they do not provide a better result than the DC method. When the number of image epochs is small the PM and PD methods may not work, but the DC method works well even if there are only 2 or 3 epochs. Comparing to the current inventory work, this approach will dramatically reduce the needs of human work and increase the efficiency of the inventory work.

The use of Multitemporal images can be extended to check the number of rice season of a year by investigating an annual temporal profile. If the pattern of temporal profile can be liked with other crop knowledge this technology would then be possibly used to recognize various crops. It would also be a very interesting topic that if the temporal relationships are implemented in a knowledge-based classification of satellite images.

Acknowledgments
The authors are appreciated that this research project was sponsored by the National Science Council of the Republic of China under the grants of NSC 86-2221-e006-007. We also would like to thanks the crop Beresu of Taiwan for providing us the test data.

Reference

[Argialas and Harlow,1990] Argialas, D.P.,and C.A. Harlow,1990. Computational Image Interpretation Models: An Overview and a Perspective, Photogrammetric and Remote Sensing,Vol,56, no6,pp.871-886.
[Brisco & Brown,1995] Brisci,B and R.J.Brown,1995. Mltidate SAR/TM Synergism for Crop classification in Western Canada , Photogrammetric Engineering and Remote Sensing, Vol 61, no.8,pp. 1009-1112
[Derenyi & Tuerker,1996]Derenyi E. and M.Tuerker, 1996. polygon Based of remotely sensed images in an Intregrated Geographic Information System ,international and Remote archives o photogrammetry and remote Sensing Vol.31, Part B4, Commission IV,pp.212-215.
[Huang et.al.,1984] Huang ,T.L.,YL.Liao, and Z.F.Her,1984. The application of Remote Sensing Technologies in the Investigation of Agricultural Resources and Land Use Classification, Agricultural &Forestry Aerial Survey Institute,Taiwan Forestry Bureau Bulletin No.10.
[Huang et al.,1985] Huang ,T.L.,Y.L.Liao, and S.C.Wang, 1985. Study on the Spectral Characteristics of Rice Crops,Agricultural & Forestry Aerial Survey Institute, Taiwan Forestry Bereau, bulltin No.12.
[Johansson, 1994] Johansson, K.,1994. Segment-Based Land-Use Clasification from SPOT Satellite Data, Photogrammetirc Engineering and Remote Sensing, Vol.60, No.1,pp.47-53.
[Johansson, 1996] Johansson, K.,1996. Generalization of Image Data to GIS Polygons for Change Detection and Data Base Revision International Archies of photogrammetry and remote Sensing Vol.31.Part B4,Commission IV,pp,384-389.
[Lo et al., 1986] Lo,T.H.,F.L.Scarpace, and T.M.Lillesand 1986. Use of Multitemporal Spectral Profile in Agricultural Land-cover Classification, photogrammetric Engineering and Remote Sensing, Vol.52,No.4,pp.535-544.
[Wolter et al.,1995] Wolter,P.T.,D.J. Mladenoff, G.E.Host, and T.R. Crow,1995. ImprovingForest Classification in the Northern Lake Using Multi-Temporal Landast Imagery, Photogrammetria Engineering and Remote Sensing, Vol.61,No.9,pp.1129-1143.

Page 3 of 3
| Previous |