Mahdis Yarmohamadi; Ali Asghar Alesheikh; Mohammad Sharif
Abstract
Dust storms are natural disasters that have severely affected human life and the environment. The majority of research in dust storm has been dedicated to the forecasting of storm-prone areas. However, developing models to predict the movement of these storms plays a significant role in the prevention ...
Read More
Dust storms are natural disasters that have severely affected human life and the environment. The majority of research in dust storm has been dedicated to the forecasting of storm-prone areas. However, developing models to predict the movement of these storms plays a significant role in the prevention and management of dust storms, because they reveal the transport pathway and identify the next vulnerable areas against the storm. In this research, a hybrid convolutional neural network (CNN) model has been developed to predict the path of dust storms based on airborne optical depth (AOD) data of MERRA-2 product for the next 12 hours. 40 storm events including 2489 storm hours in a dry region in Central and South Asia have been used for training the model. The results show that the proposed model provides an accurate prediction of the storm's path, so that for the time steps of 3, 6, 9, and 12 hours, the overall accuracy values are 0.9806, 0.9810, 0.9813, and 0.9790, respectively, the F1 score values are 0.8490, 0.8524, 0.8530, and 0.8384, respectively, and the Kappa coefficient values are 0.8387, 0.8424, 0.8431, and 0.8273, respectively.
Behrooz Moradi; AbasAli Mehraban; Morteza Mohammadi
Abstract
Autonomous landing is a key challenging in the domain of UAV navigation systems. Developing an autonomous landing system requires a precise estimation of the UAV pose relative to landing marker, particularly in vision systems this involves precise Helipad recognition. It seems that traditional approaches ...
Read More
Autonomous landing is a key challenging in the domain of UAV navigation systems. Developing an autonomous landing system requires a precise estimation of the UAV pose relative to landing marker, particularly in vision systems this involves precise Helipad recognition. It seems that traditional approaches including cascade classifiers, image matching and segmentation techniques to have major challenges in different weather conditions and scales. On the other hand, convolutional neural networks (CNNs) have been introduced as a powerful tool in the visual recognition systems in the recent years but the high computational cost of this techniques, limited their performance in the low cost and light weight UAVs. The aim of this paper is to compare the convolutional neural networks and cascade classifier for helipad detection. The results show that CNNs are invariant under translation, rotation, scaling and occlusion. The detection accuracy of this method is 99.1 % which is 3 % more than cascade classifier while its running time is suitable for real time UAV applications.
Mohammad Reza Gili; Davoud Ashourloo; Hossein Aghighi; Ali Akbar Matkan; Alireza Shakiba
Abstract
Changes in crop growth at relatively short intervals, asymmetry of cultivation of similar crops, the spectral similarity between different crops at certain times of the growing season, and lack of ground data make classifying crops in satellite imagery a challenging task. Changing the amount of canopy ...
Read More
Changes in crop growth at relatively short intervals, asymmetry of cultivation of similar crops, the spectral similarity between different crops at certain times of the growing season, and lack of ground data make classifying crops in satellite imagery a challenging task. Changing the amount of canopy and greenness during the growing season is one of the most prominent characteristics of vegetation, including agricultural products, which can be monitored by using time series of vegetation indices that have useful information about the sequence of phenological features of crops. The use of deep learning methods with the ability of learning sequential information obtained from these time series can be useful in crop mapping and reducing dependence on ground data. The LSTM network is one of the types of RNNs in sequential data analysis that has the ability to learn long-term sequences of time-series information. Therefore, in this study, after extracting the NDVI time-series of 9 different dates from Sentinel-2 satellite images for a region located in Moghan plain, with ground labeled data related to the type of crops cultivated, we trained a convolutional LSTM network. Then we used this trained network to classify agricultural products in another region of the plain as a test site, and achieved an overall accuracy of 82% and a kappa coefficient of 0.8. Increasing the number of ground samples and selecting the exact boundary of crops, can increase the efficiency of the method used.
Nacer Farajzadeh; Hiwa Ebrahimzadeh
Abstract
The development of automatic road and building detection systems in aerial imagery are always faced with challenges such as the appearance of buildings, illumination changes, imaging angles, and the density of roads and buildings in urban areas, to name a few. In recent years, employing multi-layered ...
Read More
The development of automatic road and building detection systems in aerial imagery are always faced with challenges such as the appearance of buildings, illumination changes, imaging angles, and the density of roads and buildings in urban areas, to name a few. In recent years, employing multi-layered approach in artificial neural networks, known as deep neural networks, has attracted many researchers in this field (and the other fields alike), achieving stunning results. However, the use of fully connected layers in this approach, significantly increases the average processing time and results in an overfitted model. In addition, in most of these methods, a single-class approach has been considered. That is, detecting the roads and the buildings from natural scenes is not possible at the same time, and therefore, it is necessary to build separate binary models for each of them. The main goal of this research is to design a new architecture by which the produced model can be able to simultaneously detect roads and buildings from natural scenes, and thus minimizing the complexity of the classification process. In addition, in the proposed architecture, excluding all fully connected layers from the traditional multi-layered architectures is considered in order to reduce the average processing time. The results of the experiments performed on the Massachusetts dataset, show that the proposed architecture performs 38% faster than the other deep neural network-based methods, and also increases the accuracy by an average of 2%.
nima farhadi; Abas Kiani; Hamid Ebadi
Abstract
Object detection is one of the fundamental issues in image interpretation process, especially from remote-sensing imagery. One of the most effective and efficient methods in this field is the use of deep learning algorithm for feature extraction and interpretation. An object is a collection of unique ...
Read More
Object detection is one of the fundamental issues in image interpretation process, especially from remote-sensing imagery. One of the most effective and efficient methods in this field is the use of deep learning algorithm for feature extraction and interpretation. An object is a collection of unique patterns that differ with own adjacent properties. This difference usually occurs in one or more features simultaneously, which can be indicated by the difference in shape, color, and gray values. In this regard, the use of deep learning as an efficient branch of machine learning can be useful in generating high-level concepts through learning in different layers. In this research, a database based on the environmental and geographical conditions from some Iranian airports was created. Additionally, an optimal learner model was developed with a convolutional neural network. For this purpose, in the raw data processing section, besides using the transfer learning method, some vectors were extracted to classify the objects and delivered to an SVM model. The output values were compared with the values obtained from the test image for each object, and they were analyzed in a repeatable process for structural matching. Precision of 98.21% and F1-Measure of 99.1% was achieved, for identification of the target objects