mahdi naderi
Abstract
In recent decades, land use and land cover changes information has been successfully derived from remote sensing data at various levels, from local to global scale. Accurate and frequent monitoring of these changes is required for urban planning and sustainable management of land resources. In this study, ...
Read More
In recent decades, land use and land cover changes information has been successfully derived from remote sensing data at various levels, from local to global scale. Accurate and frequent monitoring of these changes is required for urban planning and sustainable management of land resources. In this study, an object-oriented approach using a combination of GLCM, SNIC, and machine learning algorithms is presented to classify the LULC of a part of the lands of North Mahabad, West Azerbaijan, in 2019 using satellite images in Google Earth Engine. For this purpose, after preparing the initial dataset, which contains the bands of Sentinel-1 and Sentinel-2 images, the ALOS digital surface model, and NDVI, BSI, SAVI, and total scattering power indices, two pixel-based and object-oriented approaches, as well as the random forest algorithm, were used to classify land use and land cover, and their results were compared to explain the best approach in terms of the accuracy of the various classes. In the object-oriented approach, textural measures were extracted by applying the GLCM matrix to the initial dataset. Due to the increase in the number of bands, the PCA method was used to reduce the dimensions of the image. Finally, by combining the segmentation layer obtained from the SNIC algorithm and the PC1 layer, the random forest algorithm was considered to produce land use and land cover maps of the study area. According to the research findings, the object-oriented approach performed better than the pixel-based approach in classifying various land use classes in the study area, with an overall accuracy and kappa coefficient of 86.40% and 0.8307, respectively, compared to 82.73% and 0.8028. The results of the accuracy evaluation criteria showed that the producer accuracy of most of the classes except for corn, fall irrigated vegetables, and wheat, and barley irrigated in the object-oriented approach was higher than the pixel-based method, and their classification accuracy was more than 90%. Additionally, water, build-up, corn, and sugar beet classes have the highest user accuracy in the object-oriented LULC map. The findings showed that the appropriate determination of the super-pixel size of the SNIC clustering algorithm and the use of GLCM texture criteria effectively improved the performance of the proposed approach in land use and land cover classification.
Nastaran Nazariani; Asghar Fallah; Habibolah Ramezani Moziraji; Hamed Naghavi; Hamid Jalilvand
Abstract
Gathering accurate information for statistics requires high cost and precision. The time factor is also one of the important issues that should be seriously considered in statistics. Therefore, the use of sampling methods and satellite images will be a good alternative for this purpose. In the present ...
Read More
Gathering accurate information for statistics requires high cost and precision. The time factor is also one of the important issues that should be seriously considered in statistics. Therefore, the use of sampling methods and satellite images will be a good alternative for this purpose. In the present study, the aim of the effect of different cluster sampling schemes in estimating the quantitative characteristics of the traditional forests of Olad Ghobad in Koohdasht township, Lorestan province using Sentinel 2 sensor images. To estimate the studied characteristics, 150 clusters in the form of six designs (triangular, square, star 1, linear, L-shaped, star 2) were implemented in the region. Then, in each subplot, the characteristics of the number and area of the tree canopy were measured. Afterimage preprocessing and appropriate image processing (principal component analysis, texture analysis, and different spectral ratios to create important plant indices), the corresponding digital values of the ground sample plots are extracted from the spectral bands and used as independent variables. Modeling was performed using nonparametric methods of random forest, support vector machine, and nearest neighbor. The results showed that the average density per hectare was 51 and the canopy area was 32.94%. The diagram of the mean squares of the error of the training and test data against the number of trees for the characteristic number per hectare and canopy showed that the optimal number of trees was obtained at approximately 75 and 350 points. The results of validation according to the percentage of squared mean squared error showed that for both density and canopy surface characteristics of random forest algorithm with linear and double star sampling designs with the squared percentage of mean squared error respectively (46.00%) and (10.44%) and Bias (-0.02%, 2.82%) along with cluster sampling designs linear and double star, respectively, had better performance in modeling. In general, the results showed that the use of different cluster sampling schemes, nonparametric modeling methods, and Sentinel2 sensor images can better performance estimate the quantitative characteristics of Zagros forests.
Reza Shahhoseini; Kamal Azizi; Arastou Zarei; Fatemeh Moradi
Abstract
Land use maps describe the spatial distribution of natural resources, cultural landscapes, and human settlements that are essential for decision-makers. Therefore, the accuracy of maps obtained from the classification of satellite images is very effective in uncertainty for urban management. Due to the ...
Read More
Land use maps describe the spatial distribution of natural resources, cultural landscapes, and human settlements that are essential for decision-makers. Therefore, the accuracy of maps obtained from the classification of satellite images is very effective in uncertainty for urban management. Due to the uniform quality of images in large areas at regular intervals, remote sensing images are essential for land use maps. The primary purpose of this study is to present a proposed method to create an accurate land cover map in urban areas using a combination of Sentinel-1 and Sentinel-2 data. For this purpose, the features of the backscattering coefficient VV and the two parameters obtained from the H-α decomposition method (entropy, alpha) of Sentinel-1 radar images and the features of the blue, green, red band, NDVI, NDWI, MNDWI, and SWI were extracted from Sentinel-2 Multispectral images and used as influential components to classify the urban area. To separate agricultural areas from other coatings, the SWI index was used. Elevation data have also been used to optimally distinguish complex classes with different topographies. We evaluated the extraction of effective indicators from these two datasets in an object-oriented approach based on support vector machine algorithms and random forest for land use classification. The results showed that using properties extracted from radar and Multispectral images simultaneously in the object-oriented classification method could altogether determinate the object's properties in the study area. When optical and radar data were used simultaneously for both classification algorithms, the overall accuracy classification increased. For the stochastic forest method, which provided the highest accuracy, the overall accuracy for the radar and optics data combination approach increased by 13% and 5%, respectively, compared to the radar feature approach and the optics feature approach alone. There was also a significant difference in classification accuracy at all levels between the support vector machine classification algorithm and the random forest. The results showed that the random forest classification method's overall accuracy and support vector machines were 83.3 and 79.8%, respectively, and the kappa coefficient was 0.72 and 0.68%, respectively.
Ardalan Daryaei; Hormoz Sohrabi; Clement Atzberger; Markus Immitzer
Abstract
Despite the low area coverage, riparian vegetation presents several ecosystem services. But there is no precise spatial information on these ecosystems in Iran. Considering the lack of such information, mapping and providing a spatial database is essential. Due to the mixture of these vegetation types ...
Read More
Despite the low area coverage, riparian vegetation presents several ecosystem services. But there is no precise spatial information on these ecosystems in Iran. Considering the lack of such information, mapping and providing a spatial database is essential. Due to the mixture of these vegetation types and other land covers, the detection of these types of vegetation is challenging, and choosing an appropriate classification method is of great significance. This study examines the performance of pixel-based and object-based classification approaches for the detection of these vegetation types using freely available Sentinel-2 imagery. Five different riparian areas in Chaharmahal-va-Bkhtiari province were selected and used for training the classification models. We used random forest algorithm and multi-temporal Sentinel-2 data to perform the classification models. The validation of classification models was based on independent validation points spread across the province. Our results showed that multi-temporal Sentinel-2 imagery has a high capability to map riparian vegetation in the Zagros Mountains. Moreover, the pixel-based classification approach (overall accuracy of 83.9%) represents more accurate results compared to the object-based classification approach (overall accuracy of 76.7%). Overall, this study recommends a pixel-based classification approach to map these vegetation types using freely available multi-temporal Sentinel-2 imagery. Note that it is important to use pure pixels for training the classification models.
Zeinab Ghodsi; Mir Masoud Kheirkhah Zarkesh; Bagher Ghermezcheshmeh
Abstract
Land-cover/land-use maps are necessary for monitoring land changes and proper planning for managers in agriculture, natural resources and environment fields each year. The method of field data collection using GPS and land survey is time-consuming and costly. Therefore satellite images which have entire ...
Read More
Land-cover/land-use maps are necessary for monitoring land changes and proper planning for managers in agriculture, natural resources and environment fields each year. The method of field data collection using GPS and land survey is time-consuming and costly. Therefore satellite images which have entire coverage and repetition of collection, low cost and real-time data, are usually used so that land-cover/land-use maps are produced. Accurate mapping using technique suitable for today is a key factor. Although in the past, conventional classification methods have been applied to images such as Landsat, using new satellite images and modern classifiers specially machine learning has been growing recently and their effectiveness in preparing land-cover/land-use maps has been very successful. Another advantage of satellite images is repetitious collection and according to that, vegetation changes through time can be used to differentiate land cover types. The Sentinel-2 satellite with the superiority of a pixel rating of 10 meters is one of the appropriate tools to discriminate land cover types. In the current study, Support Vector Machine and Random Forest classifiers on multi-temporal Sentinel-2 images were used to differentiate land use and crop types of Sanjabi plain in Ravansar and their accuracies were compared. To do so, after sampling, Principal Component Analysis was performed for four dates in crops’ growing season and PC1,2,3 bands of the images were combined. The two techniques were implemented on the layerstacks of PC1,2,3 bands of the images and the training samples. Results of accuracy assessments showed that Support Vector Machine, with overall accuracy of 91.36% and Kappa coefficient of 0.8927, produces a more precise land use and crop map rather than Random Forest method.
A Shamsoddini; Sh Esmaeili
Volume 9, Issue 2 , December 2017, , Pages 117-132
Abstract
Differentiating agricultural areas which are not covered by vegetation from bare lands as well as identifying bare lands from urban areas in medium spatial resolution images, e.g. Landsat imagery, are usually difficult and erroneous tasks which lead to the inaccurate classification results. Therefore, ...
Read More
Differentiating agricultural areas which are not covered by vegetation from bare lands as well as identifying bare lands from urban areas in medium spatial resolution images, e.g. Landsat imagery, are usually difficult and erroneous tasks which lead to the inaccurate classification results. Therefore, this study aims to present a new approach to increase the accuracy of the classification. For this purpose, different scenarios were applied based on different input attributes. The input attributes comprised of spectral bands, textural attributes, i.e. grey level co-occurrence matrix (GLCM), and two types of indices including spatial and thermal attributes proposed in this study. Three classification methods, maximum likelihood (ML), artificial neural networks (ANN), and support vector machine (SVM) embedded with different kernels, were applied to examine different scenarios. The results showed that SVM algorithm embedded with Radial basis function (RBF) results in better accuracy, with overall accuracy of 98.81% and Kappa coefficient of 98.25%, when all types of input attributes were combined together. Finally, the variable importance analysis by random forest feature selection indicated that the proposed indices played an important role to execute more efficient classification by SVM.
