scholarly journals Evaluating the Performance of a Random Forest Kernel for Land Cover Classification

2019 ◽  
Vol 11 (5) ◽  
pp. 575 ◽  
Author(s):  
Azar Zafari ◽  
Raul Zurita-Milla ◽  
Emma Izquierdo-Verdiguier
Keyword(s):  
Random Forest ◽  
Land Cover ◽  
Satellite Image ◽  
Vegetation Indices ◽  
Computational Cost ◽  
High Dimensional ◽  
Support Vector ◽  
Spectral Features ◽  
Salinas Valley ◽  
Rbf Kernel

The production of land cover maps through satellite image classification is a frequent task in remote sensing. Random Forest (RF) and Support Vector Machine (SVM) are the two most well-known and recurrently used methods for this task. In this paper, we evaluate the pros and cons of using an RF-based kernel (RFK) in an SVM compared to using the conventional Radial Basis Function (RBF) kernel and standard RF classifier. A time series of seven multispectral WorldView-2 images acquired over Sukumba (Mali) and a single hyperspectral AVIRIS image acquired over Salinas Valley (CA, USA) are used to illustrate the analyses. For each study area, SVM-RFK, RF, and SVM-RBF were trained and tested under different conditions over ten subsets. The spectral features for Sukumba were extended by obtaining vegetation indices (VIs) and grey-level co-occurrence matrices (GLCMs), the Salinas dataset is used as benchmarking with its original number of features. In Sukumba, the overall accuracies (OAs) based on the spectral features only are of 81.34 % , 81.08 % and 82.08 % for SVM-RFK, RF, and SVM-RBF. Adding VI and GLCM features results in OAs of 82 % , 80.82 % and 77.96 % . In Salinas, OAs are of 94.42 % , 95.83 % and 94.16 % . These results show that SVM-RFK yields slightly higher OAs than RF in high dimensional and noisy experiments, and it provides competitive results in the rest of the experiments. They also show that SVM-RFK generates highly competitive results when compared to SVM-RBF while substantially reducing the time and computational cost associated with parametrizing the kernel. Moreover, SVM-RFK outperforms SVM-RBF in high dimensional and noisy problems. RF was also used to select the most important features for the extended dataset of Sukumba; the SVM-RFK derived from these features improved the OA of the previous SVM-RFK by 2%. Thus, the proposed SVM-RFK classifier is as at least as good as RF and SVM-RBF and can achieve considerable improvements when applied to high dimensional data and when combined with RF-based feature selection methods.

scholarly journals Land Cover Classification of Nine Perennial Crops Using Sentinel-1 and -2 Data

2019 ◽  
Vol 12 (1) ◽  
pp. 96 ◽  
Author(s):  
James Brinkhoff ◽  
Justin Vardanega ◽  
Andrew J. Robson
Keyword(s):  
Land Cover ◽  
Large Scale ◽  
Satellite Image ◽  
Resource Planning ◽  
Support Vector ◽  
Perennial Crop ◽  
Perennial Crops ◽  
Object Based ◽  
Rbf Kernel

Land cover mapping of intensive cropping areas facilitates an enhanced regional response to biosecurity threats and to natural disasters such as drought and flooding. Such maps also provide information for natural resource planning and analysis of the temporal and spatial trends in crop distribution and gross production. In this work, 10 meter resolution land cover maps were generated over a 6200 km2 area of the Riverina region in New South Wales (NSW), Australia, with a focus on locating the most important perennial crops in the region. The maps discriminated between 12 classes, including nine perennial crop classes. A satellite image time series (SITS) of freely available Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 multispectral imagery was used. A segmentation technique grouped spectrally similar adjacent pixels together, to enable object-based image analysis (OBIA). K-means unsupervised clustering was used to filter training points and classify some map areas, which improved supervised classification of the remaining areas. The support vector machine (SVM) supervised classifier with radial basis function (RBF) kernel gave the best results among several algorithms trialled. The accuracies of maps generated using several combinations of the multispectral and radar bands were compared to assess the relative value of each combination. An object-based post classification refinement step was developed, enabling optimization of the tradeoff between producers’ accuracy and users’ accuracy. Accuracy was assessed against randomly sampled segments, and the final map achieved an overall count-based accuracy of 84.8% and area-weighted accuracy of 90.9%. Producers’ accuracies for the perennial crop classes ranged from 78 to 100%, and users’ accuracies ranged from 63 to 100%. This work develops methods to generate detailed and large-scale maps that accurately discriminate between many perennial crops and can be updated frequently.

scholarly journals Sentinel-2 Satellite Imagery for Urban Land Cover Classification by Optimized Random Forest Classifier

2021 ◽  
Vol 11 (2) ◽  
pp. 543
Author(s):  
Tianxiang Zhang ◽  
Jinya Su ◽  
Zhiyong Xu ◽  
Yulin Luo ◽  
Jiangyun Li
Keyword(s):  
Random Forest ◽  
Land Cover ◽  
Satellite Imagery ◽  
Parameter Optimization ◽  
Satellite Image ◽  
Land Cover Classification ◽  
Random Forest Classifier ◽  
Support Vector ◽  
Wide Range ◽  
Sentinel 2

Land cover classification is able to reflect the potential natural and social process in urban development, providing vital information to stakeholders. Recent solutions on land cover classification are generally addressed by remotely sensed imagery and supervised classification methods. However, a high-performance classifier is desirable but challenging due to the existence of model hyperparameters. Conventional approaches generally rely on manual tuning, which is time-consuming and far from satisfying. Therefore, this work aims to propose a systematic method to automatically tune the hyperparameters by Bayesian parameter optimization for the random forest classifier. The recently launched Sentinel-2A/B satellites are drawn to provide the remote sensing imageries for land cover classification case study in Beijing, China, which have the best spectral/spatial resolutions among the freely available satellites. The improved random forest with Bayesian parameter optimization is compared against the support vector machine (SVM) and random forest (RF) with default hyperparameters by discriminating five land cover classes including building, tree, road, water, and crop field. Comparative experimental results show that the optimized RF classifier outperforms the conventional SVM and the RF with default hyperparameters in terms of accuracy, precision, and recall. The effects of band/feature number and the band usefulness are also assessed. It is envisaged that the improved classifier for Sentinel-2 satellite image processing can find a wide range of applications where high-resolution satellite imagery classification is applicable.

scholarly journals WorldView-2 Satellite Image Classification using U-Net Deep Learning Model

2021 ◽  
Vol 5 (2) ◽  
pp. 502-509
Author(s):  
Ilyas Ilyas ◽  
Lalu Muhamad Jaelani ◽  
Muhammad Aldila Syariz ◽  
Husnul Hidayat
Keyword(s):  
Support Vector Machine ◽  
Deep Learning ◽  
Maximum Likelihood ◽  
Random Forest ◽  
Land Cover ◽  
Satellite Image ◽  
Support Vector ◽  
Classification Methods ◽  
Land Cover Map ◽  
Land Cover Maps

Land cover maps are important documents for local governments to perform urban planning and management. A field survey using measuring instruments can produce an accurate land cover map. However, this method is time-consuming, expensive, and labor-intensive. A number of researchers have proposed using remote sensing, which generates land cover maps using an optical satellite image with various statistical classification procedures. Recently, artificial intelligence (AI) technology, such as deep learning, has been used in multiple fields, including satellite image classification, with satisfactory results. In this study, a WorldView-2 image of Terangun in Aceh Province, which was acquired on Aug 2, 2016, was classified using a commonly used deep-learning-based classification, namely, U-net. There were eight classes used in the experiment: building, road, open land (such as green open space, bare land, grass, or low vegetation), river, farm, field, aquaculture pond, and garden. For comparison, three classification methods: maximum-likelihood, random forest, and support vector machine, were performed compared to U-Net. A land cover map provided by the government was used as a reference to evaluate the accuracy of land cover maps generated using two classification methods. The results with 100 randomly selected pixels revealed that U-Net was able to obtain a 72% and 0.585 for overall and kappa accuracy, respectively; whereas, overall accuracy and kappa accuracy for the maximum likelihood, random forest and support vector machine methods were  49% and 0.148; 59% and 0.392; and 67% and 0. 511; respectively. Therefore, U-Net outperformed those three of classification methods in classifying the image.  

scholarly journals High-Resolution U-Net: Preserving Image Details for Cultivated Land Extraction

Sensors ◽  
2020 ◽  
Vol 20 (15) ◽  
pp. 4064
Author(s):  
Wenna Xu ◽  
Xinping Deng ◽  
Shanxin Guo ◽  
Jinsong Chen ◽  
Luyi Sun ◽  
...  
Keyword(s):  
High Resolution ◽  
Random Forest ◽  
Land Cover ◽  
Image Texture ◽  
Support Vector ◽  
Spectral Variation ◽  
Cultivated Land ◽  
Resource Monitoring ◽  
K Nearest Neighbors ◽  
Image Details

Accurate and efficient extraction of cultivated land data is of great significance for agricultural resource monitoring and national food security. Deep-learning-based classification of remote-sensing images overcomes the two difficulties of traditional learning methods (e.g., support vector machine (SVM), K-nearest neighbors (KNN), and random forest (RF)) when extracting the cultivated land: (1) the limited performance when extracting the same land-cover type with the high intra-class spectral variation, such as cultivated land with both vegetation and non-vegetation cover, and (2) the limited generalization ability for handling a large dataset to apply the model to different locations. However, the “pooling” process in most deep convolutional networks, which attempts to enlarge the sensing field of the kernel by involving the upscale process, leads to significant detail loss in the output, including the edges, gradients, and image texture details. To solve this problem, in this study we proposed a new end-to-end extraction algorithm, a high-resolution U-Net (HRU-Net), to preserve the image details by improving the skip connection structure and the loss function of the original U-Net. The proposed HRU-Net was tested in Xinjiang Province, China to extract the cultivated land from Landsat Thematic Mapper (TM) images. The result showed that the HRU-Net achieved better performance (Acc: 92.81%; kappa: 0.81; F1-score: 0.90) than the U-Net++ (Acc: 91.74%; kappa: 0.79; F1-score: 0.89), the original U-Net (Acc: 89.83%; kappa: 0.74; F1-score: 0.86), and the Random Forest model (Acc: 76.13%; kappa: 0.48; F1-score: 0.69). The robustness of the proposed model for the intra-class spectral variation and the accuracy of the edge details were also compared, and this showed that the HRU-Net obtained more accurate edge details and had less influence from the intra-class spectral variation. The model proposed in this study can be further applied to other land cover types that have more spectral diversity and require more details of extraction.

A bisection-sampling-based support vector regression–high-dimensional model representation metamodeling technique for high-dimensional problems

2016 ◽  
Vol 231 (12) ◽  
pp. 2173-2186 ◽  
Author(s):  
Yaping Ju ◽  
Geoff Parks ◽  
Chuhua Zhang
Keyword(s):  
Support Vector Regression ◽  
Computational Cost ◽  
Model Representation ◽  
High Dimensional ◽  
Support Vector ◽  
High Dimensional Model Representation ◽  
Dimensional Model ◽  
Comparison Results ◽  
Modeling Accuracy ◽  
Metamodeling Technique

A major challenge of metamodeling in simulation-based engineering design optimization is to handle the “curse of dimensionality,” i.e. the exponential growth of computational cost with increase of problem dimensionality. Encouragingly, it has been reported recently that a high-dimensional model representation assisted by a radial basis function is capable of deriving high-dimensional input–output relationships at dramatically reduced computational cost. In this article, support vector regression is employed as an alternative to be coupled with high-dimensional model representation for the metamodeling of high-dimensional problems. In particular, the bisection sampling method is proposed to be used in the metamodeling process to generate high-quality training samples. Testing and comparison results show that the developed bisection-sampling-based support vector regression–high-dimensional model representation metamodeling technique can achieve high modeling accuracy with a smaller number of training sample evaluations. For the problem examined in this study, the bisection-sampling-based support vector regression–high-dimensional model representation enables high modeling accuracy and linear computational complexity as the problem dimensionality increases. Analysis of this performance advantage shows that the use of bisection method enables the developed metamodeling technique to be more effective in dealing with high-dimensional problems.

scholarly journals Landslide Scars Detection using Remote Sensing and Pattern Recognition Techniques: Comparison Among Artificial Neural Networks, Gaussian Maximum Likelihood, Random Forest, and Support Vector Machine Classifiers

2020 ◽  
Vol 72 (4) ◽  
pp. 665-680
Author(s):  
Tatiana Dias Tardelli Uehara ◽  
Sabrina Paes Leme Passos Corrêa ◽  
Renata Pacheco Quevedo ◽  
Thales Sehn Körting ◽  
Luciano Vieira Dutra ◽  
...  
Keyword(s):  
Remote Sensing ◽  
Pattern Recognition ◽  
Support Vector Machine ◽  
Maximum Likelihood ◽  
Random Forest ◽  
Risk Mitigation ◽  
Vegetation Indices ◽  
Support Vector ◽  
Visual Interpretation ◽  
Artificial Neural

Landslide inventory is an essential tool to support disaster risk mitigation. The inventory is usually obtained via conventional methods, as visual interpretation of remote sensing images, or semi-automatic methods, through pattern recognition. In this study, four classification algorithms are compared to detect landslides scars: Artificial Neural Network (ANN), Maximum Likelihood (ML), Random Forest (RF) and Support Vector Machine (SVM). From Sentinel-2A imagery and SRTM’s Digital Elevation Model (DEM), vegetation indices and slope features were extracted and selected for two areas at the Rolante River Catchment, in Brazil. The classification products showed that the ML and the RF presented superior results with OA values above 92% for both study areas.  These best accuracy’s results were identified in classifications using all attributes as input, so without previous feature selection.

Sign in / Sign up

Export Citation Format

Share Document