| Tobacco is a kind of special commodity which is regulated by the National Government.Tobacco company is an integration of enterprises and is high degree of monopoly. According to the relevant provisions, the tobacco planting unified management by the tobacco company.Tobacco field area measurement is an important means to monitor farmer actual transplanting number and area of tobacco are in conformity with the contract signed number and area.Remote sensing yield estimation can timely and accurately grasp the dynamic information of large area of tobacco production. It is a kind of no contact, no failure damage, simple, easy,time-sensitive and low cost method. This is the basis on which tobacco company could timely and accurately grasp the situation of tobacco production, adjust the tobacco planting plan and evaluate how policies are implemented. It can lay the foundation for the scientific research and quantization management in tobacco.Traditional field measurements are difficult to monitor the mountain area of relay-cropping tobacco field in hilly regions, a fast, high precision estimation approach is developed to solve this problem. This article takes Yishui County of Linyi city in Shandong province as an example. Use ENVI5.1 to process the ZY-3 panchromatic image of 2.5 m and multi-spectral image of 5.8 m. Import 1:10000 DEM data during orthorectification to banishing or limiting projection error at hill area. Get 200 typical ground control points from field investigation. Estimate the area of relay-cropping tobacco field with the method of object-oriented classification by building the rules of GCP’s spectrum, texture and shape attribute, after segmentation and merging image intelligent. Through the analysis of the multispectral image information to calculate the vegetation index. Build prediction regression model of tobacco multispectral information- tobacco yield. Select the appropriate model based on the evaluation results.The conclusion of this paper is as follows:(1)Remote sensing image preprocessing can eliminate image imaging problem. Using1:10000 DEM data in Orthographical correction can correct the image error caused by terrain fluctuation well. Gram-Schmidt method turn out to have a very good result, not only showing the advantages of high resolution but also keeping the authenticity of spectral values.(2)The characteristic feature extracting effect of object-oriented approach is much better than unsupervised classification and decision tree classification. Object-oriented classification method can significantly improve the classification accuracy by taking full use of image’sspectrum, shape, structure, texture, layout and the context information between image features.Image segmentation and merging is the precondition of object-oriented characteristic feature extraction. The principle that selecting the best segmentation merger scale is to make the tobacco field neither broken nor mixed with other features within one object. In this process a lot of experimental have been taken.The area of the tobacco field that extract from post-processing of image classification is 14088.46 hm2. The classification accuracy turn out to be 94.63%.(3)Analysis the correlation of tobacco bands and vegetation index with the actual production of tobacco we can learn that: band 1 has moderate correlations with tobac co yield; band 2 and band 3 have high correlations with tobacco yield; band 4, NDV I, DVI, and RVI have significant correlations with tobacco yield. NDVI and RVI hav e the best correlations with tobacco yield to the number of 0.968 and 0.967.(4)Research has established 4 kind of tobacco yield estimation model with band 4, NDVI,DVI, and RVI. Each kind of tobacco yield estimation model’s fit degree are different from others. And different equation also have different fit degree even though in one kind of tobacco yield estimation model. NDVI has the best fit degree; band4 is the worst; DVI and RVI are middling. Precision analysis was carried out on the yield estimation model. Band 4has the lowest estimating model accuracy of 86.7%. Vegetation index estimating models accuracy are all above 90%. Among vegetation index estimating models. NDVI has the high model accuracy among vegetation index estimating models. It’s model accuracy is 97.3%... |
PDF Full Text Request