ENGINEERING JOURNAL OF GEOSPATIAL INFORMATION TECHNOLOGY
Year:2022 | Volume:9 | Issue:4|Papers Count:6

Raw point clouds usually include noise and outliers. Also, the point clouds generated by photogrammetry methods are noisier than the point clouds that are derived from active methods such as laser scanners, hence many challenges for reconstructing and meshing surface using these three-dimensional data would be possible. Also, maintaining sharp features is essential during the process of noise removal. Many techniques have been developed to remove noise from the point cloud, but only a few of them are suitable for maintaining Sharp features during the noise removal process. This paper tries to provide a new statistical method with the ability to maintain sharp features, to remove noise. In the proposed method, first, the point cloud is clustered using the mean-shift clustering algorithm. As the clustering accuracy depends on the kernel size, the optimal size of the window is achieved through the hill climbing optimization. Then, in each cluster, the distance between each point and the mean of the other points of that cluster is calculated,next, appropriate thresholds are used to detect and remove noise from point cloud by applying them on the number of members of each cluster and computed distances. So the sharp features, such as the edges, are preserved. The experimental results obtained from the implementation of the proposed method on the three sets of 3D data, provided by the laser scanner, illustrate that this method, compared with the other methods presented in the literature review, increases the accuracy about 4% in noise removing and 5. 19 percent in maintaining sharp features.

Wetlands are one of the most important ecological resources. Detecting their long-term changes plays a key role in the quality of the management of such areas. These unique ecosystems in the world with high ecological diversity are threatened by various natural factors such as: decrease in rainfall, increase in temperature, increase in evaporation, drought, and so on. This research focuses on developing a practical and effective framework for long-term monitoring of water area of the wetland using parameters affecting the wetland and Landsat time series images, all obtained from the Google Earth Engine (GEE) system. In this study, in order to determine the Changes in the water body, normalized difference water index (NDWI) has been used to have a better discrimination between water and other classes of the region. The changes in the water area of Anzali Wetland and the natural factors affecting it were studied in the period of 240 months between January 2000 and December 2019. Then, by using the method based on MLP machine learning and the parameters affecting the surface changes of the wetland as the input of the network, the surface changes of the wetland with average root mean square error (RMSE) of 0. 977 were modeled. Also, in order to predict the severe surface changes of the wetland in the future, the surface changes of the wetland and all parameters for a long period (last 20 years) were examined on a monthly basis using the Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) methods. Finally, according to the results obtained from the previous stages and detecting the factors that have a greater impact on the wetland and due to uncertainty, nonlinearity of the behavior of variables, the Fuzzy Inference System (FIS) was designed to create a wetland drought warning system. Therefore, the developed model can be easily implemented to be used continuously for the management and monitoring of wetlands.

One of the necessary pieces of information for policy-making and urban management is an up-to-date land use map, while the time and cost of producing and updating spatial information using traditional mapping methods and by national or private mapping organizations are too high. The advancement of technology such as smart phones, real-time positioning, and social network development has resulted in the mass production of User Generated Geographic Content (UGGC). The purpose of this study is to identify the land use type of the parcels using UGGCs. In this research six categories of urban land use types have been taken into cosideration: residential, commercial/shopping, office/service, mixed, entertainment/recreational, and the other ones,and the social network data of Twitter is used as User-generated content. Deep learning classification and Recurrent Neural Network (RNN) are utilized to analyze the user-generated data. To eliminate the imbalance of the input data, the Support Vector Machine (SVM) algorithm is utilized. Evaluation of the results of the proposed method demonstrates classification of urban land uses with an overall accuracy of 64%. Among urban use classes, the residential one is the best with 77 percent accuracy. The area under the ROC curve is also 0. 88, which indicates the appropriate reliability of the proposed method. To eliminate data imbalance, comparing the results of the SVM algorithm with the random method of sampling, reveals that SVM presents higher accuracy.

One of the applications of remote sensing is to study and classify the alteration areas, which is one of the fastest methods to explore the porphyry copper deposit, determine its accumulation center and location of drilling points. The aim of this study is to identify argillic, phyllite, and propylitic alterations in small exploration ranges and to determine porphyry copper accumulation area as well. In this regard, an algorithm on the basis of deep convolutional cane crusts was designed. In the proposed algorithm, first, preprocessings such as geometric and spectral correction and repairing and training data amplification were performed in order to prepare RGB and SWIR data of the ASTER sensor to enter the chip. The proposed convolutional shear chip (CNN) has a coder-decoder structure that in the coding stage different and efficient features are extracted at different scales and in the decoding stage the generated features are combined to estimate the alteration regions. Then, the desired network was implemented for the images of the studied exploratory area called "Customs Mouth" located in Jiroft city, and the alteration areas of the region were extracted. For field evaluation of the results, lithological and geochemical methods were used on 84 samples. By merging the network results, extracting the geometric structure of the alterations and locating it on the fine copper and gold interpolation map of the region, and examining the lithological results, the alterations of the region with a statistical accuracy of sensitivity parameters: 0. 943, F1 score: 0. 472, IoU: 0. 896 and lithographic detection accuracy 92% and an average copper grade above 4% were identified in these areas. The digging trenches map to extract mineral deposits was obtained on the basis of the detected alterations.

This paper presents a model, the International Reference Model 2016 (IRI), in order to improve the vertical total electron content (VTEC) maps by combining Swarm observations with global positioning system (GPS) ones. The proposed model consists of two parts: the background model and the corrections. In this paper, the IRI2016 model was selected as the background model and the corrections were modeled by spherical harmonic expansion functions up to the degree and rank 15 in a Sun-fixed reference frame. In the combination of VTECs derived from Swarm and GPS, the systematic biases of Swarm satellites are considered as unknown constant parameters in each epoch of modeling. Besides, in order to take the different accuracy levels of observational groups into consideration, the Helmert variance component estimation method is used. To evaluate the proposed model, the two-dimensional combined global ionosphere maps (GIMs) are constructed on the 28th of September 2017 and the 3rd of January 2018 with 7 and 1 kp-indices values, respectively. The comparison of the combined GIM maps with the International GNSS Service (IGS) GIM maps, shows that the combined model is more compatible with IGS maps, and adding Swarm and GPS observations to the IRI2016 background model can significantly improve the IRI2016 model, especially in oceanic regions. The results show that the root mean square (RMS) and root mean square error (RMSE) maps are decreased about 19% to 45% and 43% to 67% for the day with high Kp-index and about 13% to 40% and 15% to 43% for the day with low kp-index, respectively.

Nowadays, with the advancement of technology, numerous sensors provide high spatial and spectral resolution images. So far, several methods have been proposed for hyperspectral image classification, each of which seeks to overcome a number of computational and processing challenges of hyperspectral data. The efficiency of multi-layer perceptron neural networks is greatly reduced due to the increase in the number of parameters along with the increase of the layers, which is essential in complex topics such as hyperspectral image classification. In recent years, the concept of deep learning, especially convolutional neural networks (CNN), has attracted the attention of pattern recognition researchers due to the automatic generation of features and the reduction of parameters compared to the multi-layer perceptron neural networks by sharing the parameters in each layer. The goal of the present study is to develop a convolutional neural networks (CNN in order to classify hyperspectral images. The innovation of this study is to provide a framework to use deep learning. The proposed framework includes four steps. The first step is to reduce dimension by using the sub-space method, the second step is to prepare the CNN inputs, the third step is to augment the teaching data, and the fourth step is to design the CNN architecture. Implementation of the proposed framework on the benchmark data of the University of Pavia, despite the use of a limited number of educational data, led to the classification accuracy of 98. 3%.