Showing 3 results for Supervised Classification
N. Zahedifard, S. J. Khajeddin, A. Jalalian,
Volume 8, Issue 2 (7-2004)
Abstract
Satellite data use is finding global applications because they provide repeated cover, broad information, high electromagnetic spectral resolution, and software-hardware compatibilities. This study aims to evaluate of the Landsat TM data capabilities in land-use mapping of Bazoft River basin (Chahar Mahale Bakhtiary Province). Six spectral bands of the Landsate TM were employed to produce land-use map of the Region. The date of image acquisition was May 5th, 1998. Performance of the geometric correction completed with RMSE= 1.008 pixels. Various image enhacement methods (e.g. FCC, filtering and Vegetation Indices) were used to study the different land-covers. Field investigations were carried out using a GPS, 1:50000 scale topographic map and false color composites images. Heterogeneous land-use units were studied in 62 sample sites estimating percentage of vegetation cover. A regression analysis was performed between percentage vegetation covers and vegetation indices values of NDVI, RVI, SAVI, DVI, TSAVI1, NRVI and MSAVI2. Results show that NDVI, SAVI, TSAVI1, NRVI and MSAVI2 have high correlation coefficients. But RVI, DVI and PVI have low correlation coefficients. The resulting values of vegetation cover were density sliced to produce the land-cover map. After supervised classifications and density slicing of Vegetation Indices, classifacation accuracy was assessed and, finally, land-use map of the study area was produced with Hybrid classification method. Supervised classification with maximum likelihood method was the best technique for land-use mapping in the study area the total Kappa index was %87. In general, detection of some land-use classes through single date TM data is not feasible, these include: scattered forest trees with cultivated understory, annual grasses, and fallow lands. Also TM digital data are incapable of distinguishing small and separated rural constructions or soil-covered routes.
S. J. Khajeddin, S. Pourmanafi,
Volume 11, Issue 1 (4-2007)
Abstract
To detect the rice paddis areas in Isfahan region, the IRS-1D data from PAN, LISS III and WiFS time series were used. Geometric, atmospheric, radiometric and topographic corrections were applied to various images from 2003 to 2004. Necessary preprocessing and various analyses as well as time series composite image analyses were applied and field sampling was done for appropriate times in 2003 and 2004. Image classification was applied using suitable training sites in various images. The SWIR band capabilities were useful for NDWI (Normalized Difference Water Index) to detect the rice paddies. On PAN and LISS III images, urban areas, roads, agricultural lands, non cultivated farms, rocks and brackish soils are detectable. The error matrix was calculated to assess the produced map accuracy using the ground truth data. The total classification accuracy was %91 and the Kappa index value was %89. The rice paddy areas was about 19500 ha in 2003, detected through LISS III data, and 20450 ha through WiFS data. The paddies were 21670 in 2004 through WiFS data. The results of this study confirmed that one can use the LISS III data to detect and determine the rice paddys areas with high accuracy, and WiFS data to estimate the paddies areas with acceptable accuracy.
L. Khodakarami , A. Soffianian,
Volume 16, Issue 59 (4-2012)
Abstract
Precision farming aims to optimize field-level management by providing information on production rate, crop needs, nutrients, pest/disease control, environmental contamination, timing of field practices, soil organic matter and irrigation. Remote sensing and GIS have made huge impacts on agricultural industry by monitoring and managing agricultural lands. Using vegetation indices have been widely used for quantifying net annual production on different scales. The aim of this study was to find a rapid method with acceptable precision for the identification and classification of agricultural lands under cultivation (wheat and barley, alfalfa and potatoes). We used multi-temporal AWiFS data and applied Boolean logic and unsupervised classification.
Results indicated that Boolean logic approach had a higher accuracy and precision in comparison to unsupervised classification, although it is more complicated and time consuming.
