Analytic Study on the Characteristics of Infrared Spectrum of Oil Pool Flame.

In virtue of the severity and scale of the pollution caused by oil pool flame, space remote sensing can provide us a new way of monitoring in real time the oil pool flame pollution. Space remote sensing monitoring is based on the analysis of target spectrum characteristics. Due to lack of adequate researches on the characteristics of infrared spectrum of oil pool flame, this paper carries out the analytical study on flame spectrums of several types of oil, mixed oil and other combustible objects in outdoor space by establishing all-flame infrared testing system with the spectrum range of 1～14 µm. The results show that the spectrum curves of oil pool flame of 92# gasoline, 95# gasoline, 0# diesel, aviation kerosene and lube have similar features, that there exist characteristics emission peaks at the area of certain wave lengths­H2O characteristics emission peak for 1.1, 2.4, 2.8 and 6.3 µm, CO2 characteristics emission peak for 4.2 and 4.5 µm, C­H stretching vibration emission peak for 3.4 µm, and no obvious characteristics peak for spectrum curves of 6.3 µm and above; that there is no obvious difference in the spectrum of oil pool flame among the mixtures of 92# gasoline and 0# diesel at different proportions, that the comparison of the flame spectrum of 92# gasoline with that of wood and paper shows that there appears a characteristics emission peak at 3.4 µm; that though the flame spectrum of alcohol has similar radiated emission near 3.4 µm, the proportion of its radiation intensity to that of CO2 at 4.5 µm is far less than that for the flame spectrum of 92# gasoline; that the flame spectrum of honeycomb briquette is similar to that of gray body radiation. The differences in flame spectrum among all kinds of combustible materials are closely linked to their chemical compositions and burning reaction mechanisms. Comparative analysis on the spectrum characteristics at continuous area, intermission area and flue gas area shows that C­H stretching vibration peak only exists in continuous area, which proves that the emission peak is caused by the combustible reaction of oil and gas. This result is in line with the mechanism of oil pool combustion reaction. The experimental conclusion is of great significance in the remote-sensing recognition of oil pool flame based on the analysis of spectrum characteristics.

[Hard and soft classification method of multi-spectral remote sensing image based on adaptive thresholds].

Hard and soft classification techniques are the conventional methods of image classification for satellite data, but they have their own advantages and drawbacks. In order to obtain accurate classification results, we took advantages of both traditional hard classification methods (HCM) and soft classification models (SCM), and developed a new method called the hard and soft classification model (HSCM) based on adaptive threshold calculation. The authors tested the new method in land cover mapping applications. According to the results of confusion matrix, the overall accuracy of HCM, SCM, and HSCM is 71.06%, 67.86%, and 71.10%, respectively. And the kappa coefficient is 60.03%, 56.12%, and 60.07%, respectively. Therefore, the HSCM is better than HCM and SCM. Experimental results proved that the new method can obviously improve the land cover and land use classification accuracy.

[Division of winter wheat yield estimation by remote sensing based on MODIS EVI time series data and spectral angle clustering].

Crop yield estimation division is the basis of crop yield estimation; it provides an important scientific basis for estimation research and practice. In the paper, MODIS EVI time-series data during winter wheat growth period is selected as the division data; JiangSu province is study area; A division method combined of advanced spectral angle mapping(SVM) and K-means clustering is presented, and tested in winter wheat yield estimation by remote sensing. The results show that: division method of spectral angle clustering can take full advantage of crop growth process that is reflected by MODIS time series data, and can fully reflect region differences of winter wheat that is brought by climate difference. Compared with the traditional division method, yield estimation result based on division result of spectral angle clustering has higher R2 (0.702 6 than 0.624 8) and lower RMSE (343.34 than 381.34 kg x hm(-2)), reflecting the advantages of the new division method in the winter wheat yield estimation. The division method in the paper only use convenient obtaining time-series remote sensing data of low-resolution as division data, can divide winter wheat into similar and well characterized region, accuracy and stability of yield estimation model is also very good, which provides an efficient way for winter wheat estimation by remote sensing, and is conducive to winter wheat yield estimation.

[Winter wheat area estimation with MODIS-NDVI time series based on parcel].

Several attributes of MODIS (moderate resolution imaging spectrometer) data, especially the short temporal intervals and the global coverage, provide an extremely efficient way to map cropland and monitor its seasonal change. However, the reliability of their measurement results is challenged because of the limited spatial resolution. The parcel data has clear geo-location and obvious boundary information of cropland. Also, the spectral differences and the complexity of mixed pixels are weak in parcels. All of these make that area estimation based on parcels presents more advantage than on pixels. In the present study, winter wheat area estimation based on MODIS-NDVI time series has been performed with the support of cultivated land parcel in Tongzhou, Beijing. In order to extract the regional winter wheat acreage, multiple regression methods were used to simulate the stable regression relationship between MODIS-NDVI time series data and TM samples in parcels. Through this way, the consistency of the extraction results from MODIS and TM can stably reach up to 96% when the amount of samples accounts for 15% of the whole area. The results shows that the use of parcel data can effectively improve the error in recognition results in MODIS-NDVI based multi-series data caused by the low spatial resolution. Therefore, with combination of moderate and low resolution data, the winter wheat area estimation became available in large-scale region which lacks completed medium resolution images or has images covered with clouds. Meanwhile, it carried out the preliminary experiments for other crop area estimation.

[Integration of soft and hard classifications using linear spectral mixture model and support vector machines].

This paper presents a new soft and hard classification. By analyzing the target objects in the image distribution, and calculating the adaptive threshold automatically, the image is divided into three regions: pure regions, non-target objects regions and mixed regions. For pure regions and non-target objects regions, hard classification method (support vector machine) is used to quickly extract classified results; For mixed regions, soft classification method (selective endmember for linear spectral mixture model) is used to extract the abundance of target objects. Finally, it generates an integrated soft and hard classification map. In order to evaluate the accuracy of this new method, it is compared with SVM and LSMM using ALOS image. The RMSE value of new method is 0.203, and total accuracy is 95.48%. Both overall accuracies and RMSE show that integration of hard and soft classification has a higher accuracy than single hard or soft classification. Experimental results prove that the new method can effectively solve the problem of mixed pixels, and can obviously improve image classification accuracy.

[Farmland parcel extraction based on high resolution remote sensing image].

A new method of farmland parcel extraction from high resolution remote sensing image based on wavelet and watershed segmentation was proposed in the present paper. First, classification results were used to enhance the contrast of gray-scale value of typical pixels in the original image using the high resolution remote sensing image's spectral information. Second, wavelet transform and watershed segmentation were applied to the enhanced image, then improved region merger algorithm was used to solve the problem of over-segmentation. Finally, inverse wavelet transform was taken to get the reconstructed image, then Canny operator was introduced to add the edge information, and the result of farmland parcel segmentation was obtained. To validate the proposed approach, experiments on Quickbird images were performed, we rapidly extracted the farmland parcel from the test image, and the results had a high accuracy. Despite it had a lot to do in extracting the small size parcels, on the whole the method this paper proposed had a very good robustness. Compared with the traditional methods, it had the following advantages: (1) it was an automatic extraction method, did not need too much manual intervention, and could extract the large area of farmland parcels accurately and effectively. (2) It was a very good solution to the problem of over-segmentation by using improved region merger algorithm, and improved the accuracy of the extraction. All these indicated that the proposed approach was an effective farmland parcel extraction method based on high resolution remote sensing image.