Geographic Data
Zahra Moradi; Mohammad Sadi Mesgari
Abstract
Extended Abstract-Introduction: The growing importance of housing is not hidden from anyone in terms of the profound and significant effects it has on the various social, political, and economic dimensions of countries; Therefore, accurate and reliable price estimation definitely facilitates policy-making ...
Read More
Extended Abstract-Introduction: The growing importance of housing is not hidden from anyone in terms of the profound and significant effects it has on the various social, political, and economic dimensions of countries; Therefore, accurate and reliable price estimation definitely facilitates policy-making in this field. Hundreds of factors may affect property prices in different situations as a subset of structural, spatial, and socio-economic factors. Therefore, considering these factors, property pricing should be done efficiently. Due to the complex nature of the real estate market, research has used common deep learning algorithms such as DNN, RNN, CNN, etc., but these algorithms are not very suitable for tabular data. On the other hand, the deep learning models in property pricing are also completely definite and do not take into account data uncertainty.Materials & Method: In this article, we have tried to pay attention to the tabular structure of real estate data in applying deep learning methods. The TabNet deep new architecture is used for this purpose. In addition, at the same time as the learning process, it makes feature selection fully interpretable. In this study, also using existing combination techniques, fuzzy logic is combined with deep learning algorithms to learn complex problems faster and more accurately, to overcome the shortcomings of the certainty of deep learning models and not consider the inherent uncertainty of the data in this models. In this study, using the existing combination techniques, also using spatial information system (GIS) to provide a clearer evaluation to ensure full visualization of the spatial pattern of property properties as well as the relationship between these properties and pricing and spatial variables are included in the valuation model. In order to evaluate the proposed methods, real estate data of District 5 of Tehran were used.Results & Discussion: The order and prioritization of the impact of features on the pricing of Tehran residential properties by the TabNet algorithm indicate the significant impact of spatial factors. So that in this ranking, after the area, the two spatial characteristics of latitude and longitude have the second and third ranks, respectively. Basically, latitude and longitude indicate the criteria of neighborhoods and the type and prestige of different places in the city, and the social class of different streets and neighborhoods in the city, which is clearly a factor in influencing the price. Finally, TabNet, DNN, CNN, RNN, LSTM, Autoencoder algorithms as well as XGBoost machine learning algorithms were used for the Tehran data set, and RMSE, MA and evaluation criteria were compared, which according to the criterion, a 5% improvement in accuracy was achieved by using TabNet. Finally, the RMSE of the FuzzyTabNet hybrid algorithm for Tehran data decreased by 4/65% compared to the basic TabNet algorithm. The fuzzy Autoencoder network also improved by 5/52% compared to the common Autoencoder network.
Mohsen Abedi; Mohammad SaadatSeresht; Reza Shahhoseini
Abstract
Extended Abstract
Introduction
Nowadays, updating information collected from urban areas is of great importance, since it provides the basis for many fields of study such as land cover changes and environmental studies. Remote sensing provides an opportunity to obtain information from urban areas ...
Read More
Extended Abstract
Introduction
Nowadays, updating information collected from urban areas is of great importance, since it provides the basis for many fields of study such as land cover changes and environmental studies. Remote sensing provides an opportunity to obtain information from urban areas at different levels of accuracy while widely used in various change detection applications. Detecting changes in buildings as one of the most important features in urban areas is of particular importance. Powerful and expensive processing systems are the only way to process large volume of remote sensing and photogrammetry data generated by the ever increasing number of sources to which laymen do not have access. The present study has applied deep learning methods and high computational volume of data processing in free clouds to make this possible for the public.
Materials & Methods
Two case studies have been selected in the present study. The first includes DSM and Orthophoto images captured by drones from Mashhad in 2011 and 2016. DSM and Orthophoto images in the second case study has been collected by drones from Aqda in Yazd province in 2015 and 2018. In accordance with the type of data used and high computational volume used for processing, the present study has applied fuzzy clustering method to detect buildings with a high computational speed and deep learning method to detect their changes. Object-based method and fuzzy logic theory have been used in the first step to classify features and detect buildings. In the second step, deep learning method and DSM differentiation method were also used to detect changes in buildings and evaluate results obtained from deep learning method. In the first step, buildings were detected using descriptors extracted from terrestrial and non-terrestrial features, and related decisions were made using fuzzy logic. In the second step, DSM differentiation method has applied the masks extracted from buildings in both epochs on the related DSMs to find their difference and detects changes using an elevation threshold. In deep learning method, a convolutional neural network model was trained to detect changes in buildings during both epochs. Using the DSM of buildings in both epochs and a part of their interface, the network input layers were generated for training. Changes detected in the buildings by the differentiation method were also introduced as the output layer. Following the training and introducing the entire interface in both epochs as the input layer, the trained neural network has detect changes in the buildings. The same process was performed once more using the difference between two DSMs. In other words, a single input layer was used in the network and the rest of the process was the same as before. Finally, changes detected by the neural network was compared with changes detected in the DSM differentiation method
Results & Discussion
In the first step, buildings were detected and images were classified in accordance with the fuzzy logic. The overall accuracy of the first epoch classification in Mashhad equaled 94.6% indicating higher acuracy of object-based methods as compared to pixel-based methods. The overall accuracy of first epoch in Aqda equaled 95.5%. Neural network method detected changes in buildings with an overall accuracy of 90%. In accordance with the ground truth used in network training (both using DSMs as the input layer and the difference between the epochs as the input layer), results indicated that deep learning method is highly accurate in one-dimensional convolution mode. Moreover, the second step has applied the difference between DSMs in the two epochs and thus, many areas lacking a change in height were removed in both epochs and the network was trained more appropriately and accurately.
Conclusion
Necessity of extracting features, especially urban features such as buildings and identifying their changes over time have been investigated in the present study. Due to the high computational volume of modern remote sensing and photogrammetry data and highly expensive systems required for their processing, a new method was presented in the present study to solve this problem. Considering the type of data used and the complexity of features, object-based methods were selected instead of pixel-based methods to identify features and buildings. Deep learning method was used to detect changes in buildings. The method was also compared with DSM differentiation method. A one-dimensional convolutional neural network was used in the deep learning method. Two different modes were used in the network to train and predict changes. In the first, DSMs extracted from the buildings in each epoch were used as the input layer, while in the second one, the difference between DSMs were introduced as a single input layer to the network and the network was trained in accordance with the ground truth collected from areas with and without change obtained from the DSM differentiation method. Following the training process, changes were predicted using the trained network. Much better results were obtained from the second mode in which the difference between DSMs were used.
Hossein Bagheri; Mohammad Hassan Zali
Abstract
Extended Abstract
Introduction
The concentration of particulate matters has recently increased in the metropolitan area of Tehran resulting in many severe hazards for both the environment and citizens. Particulate matters (PM) with a diameter less than 2.5 microns (PM2.5) are considered to be one ...
Read More
Extended Abstract
Introduction
The concentration of particulate matters has recently increased in the metropolitan area of Tehran resulting in many severe hazards for both the environment and citizens. Particulate matters (PM) with a diameter less than 2.5 microns (PM2.5) are considered to be one of the most dangerous types of pollution. Estimating the concentration of these particles in Tehran is challenging due to the existence of various sources of pollution and the lack of sufficient ground stations. Aerosol optical depth (AOD) data retrieved from satellite imagery can be an alternative. However, AOD are not easily convertible into surface pollution and requires the development of appropriate models such as those based on data-driven approaches and machine learning techniques. Thus, the present study seeks to create a model to estimate the concentration of PM2.5 in Tehran employing deep generative models and in-situ measurements, meteorological data, and AOD data extracted from MODIS satellite imagery. Reviewed literature has proved the ability of deep learning techniques to solve regression and classification problems. Deep learning techniques are divided into various categories, one of which is based on the generative models seeking to reconstruct the input features. In this way, high-level and efficient features can be employed to explore the relationship between PM2.5 and AOD. Thus, the present study has investigated the potential of deep generative models for estimating PM2.5 concentration from high resolution AOD data retrieved from satellite imagery.
Materials and Study Area
As a metropolitan area suffering from air pollution particularly in winters, the capital city of Iran, Tehran was selected as the study area. PM2.5, the main source of pollution in Tehran, is mainly emitted from vehicles and especially old urban public transport fleet.
Aerosol data collected by Aqua and Terra sensors of MODIS and retrieved by Multi-angle Implementation of Atmospheric Correction (MAIAC) algorithm were used in the present study. Meteorological data were obtained from the global ECMWF climate model, and the concentration of PM2.5 was measured at air quality monitoring stations. Data were collected for a time interval of January 2013 to January 2020.
Methods
The present study has investigated the potential of deep generative models used to provide an estimate of PM2.5 concentration based on satellite AOD data. To reach such an aim, three types of deep generative neural networks, deep autoencoder (DAE), deep belief network (DBN) and conditional generative adversarial network (CGAN) were developed. Moreover, the performance of deep generative modes was compared with linear regression techniques as typical models used to explore the relation between PM2.5 and AOD data. Finally, the most accurate model for the generation of high resolution (1km) PM2.5 maps from AOD data was selected based on the performance of models.
Results and Discussion
The accuracy of each developed model was evaluated using the test data and the obtained results were compared with results obtained from other basic linear regression models. Accuracy evaluation indicated that the developed deep autoencoder (DAE) combined with support vector regression led to the highest correlation (R2 = 0.69) and lowest RMSE (10.34) and MAE (7.95) and thus, can be potentially used for high resolution estimation of PM2.5 concentration. Next was the developed deep belief network which with a performance close to DAE demonstrated its potential capability to estimate PM2.5 concentration from satellite AOD data. The CGAN network acted less accurately in the estimation of PM2.5 concentration as compared to other deep generative models, but outperformed the linear regression algorithms on the test data. To sum up, findings indicated that deep generative models have outperformed classical linear regression techniques used for high resolution estimation of PM2.5 from satellite AOD data. Among the linear methods, the highest accuracy was achieved by the Lasso algorithm with an RSME of 12.14 and MAE of 9.46 on the test data which showed the significance of regularization for the improvement of performance in linear regression algorithms. Nevertheless, the accuracy of linear regression techniques was much lower than deep generative models.
Conclusion
Finally, DAE was selected as the best model for the estimation of PM2.5 concentration across the study area and high resolution maps of PM2.5 concentration were generated using the developed model. Investigating the daily PM2.5 maps generated for two days with different air quality conditions (clean and polluted) demonstrated the efficiency of the developed DAE for PM2.5 modeling.