Search

Road network extraction from remotely sensed imagery has become a powerful tool for updating geospatial databases, owing to the success of convolutional neural network (CNN) based deep learning semantic segmentation techniques combined with the high-resolution imagery that modern remote sensing provides. However, most CNN approaches cannot obtain high precision segmentation maps with rich details when processing high-resolution remote sensing imagery. In this study, we propose a generative adversarial network (GAN)-based deep learning approach for road segmentation from high-resolution aerial imagery. In the generative part of the presented GAN approach, we use a modified UNet model (MUNet) to obtain a high-resolution segmentation map of the road network. In combination with simple pre-processing comprising edge-preserving filtering, the proposed approach offers a significant improvement in road network segmentation compared with prior approaches. In experiments conducted on the Massachusetts road image dataset, the proposed approach achieves 91.54% precision and 92.92% recall, which correspond to a Mathews correlation coefficient (MCC) of 91.13%, a Mean intersection over union (MIOU) of 87.43% and a F1-score of 92.20%. Comparisons demonstrate that the proposed GAN framework outperforms prior CNN-based approaches and is particularly effective in preserving edge information.

Mapping lithological units of an area using remote sensing data can be broadly grouped into pixel-based (PBIA), sub-pixel based (SPBIA) and object-based (GEOBIA) image analysis approaches. Since it is not only the datasets adequacy but also the correct classification selection that influences the lithological mapping. This research is intended to analyze and evaluate the efficiency of these three approaches for lithological mapping in semi-arid areas, by using Sentinel-2A data and many algorithms for image enhancement and spectral analysis, in particular two specialized Band Ratio (BR) and the Independent component analysis (ICA), for that reason the Paleozoic Massif of Skhour Rehamna, situated in the western Moroccan Meseta was chosen. In this study, the support vector machine (SVM) that is theoretically more efficient machine learning algorithm (MLA) in geological mapping is used in PBIA and GEOBIA approaches. The evaluation and comparison of the performance of these different methods showed that SVM-GEOBIA approach gives the highest overall classification accuracy (OA $\approx ~93$ %) and kappa coefficient (K) of 0, 89, while SPBIA classification showed OA of approximately 89% and kappa coefficient of 0, 84, whereas the lithological maps resulted from SVM-PBIA method exhibit salt and pepper noise, with a lower OA of 87% and kappa coefficient of 0, 80 comparing them with the other classification approaches. From the results of this comparative study, we can conclude that the SVM-GEOBIA classification approach is the most suitable technique for lithological mapping in semi-arid regions, where outcrops are often inaccessible, which complicates classic cartographic work.

Money laundering has been a global issue for decades, which is one of the major threat for economy and society. Government, regulatory and financial institutions are combating it together in their respective capacity, however still billions of dollars in fines by authorities make the headlines in the news. High-speed internet services have enabled financial institutions to deliver better customer experience through multi-channel engagements, which has led to exponential growth in transactions and new avenues for laundering the money for fraudsters. Literature shows the usage of statistical methods, data mining and Machine Learning (ML) techniques for money laundering detection, but limited research on Deep Learning (DL) techniques, primarily due to lack of model interpretability and explainability of the decisions made. Several studies are conducted on application of ML for Anti-Money Laundering (AML), and Explainable Artificial Intelligence (XAI) techniques in general, but lacks the study on usage of DL techniques together with XAI. This paper aims to review the current state-of-the-art literature on DL together with XAI for identifying suspicious money laundering transactions and identify future research areas. Key findings of the review are, researchers have preferred variants of Convolutional Neural Networks, and AutoEncoder; graph deep learning together with natural language processing is emerging as an important technology for AML; XAI use is not seen in AML domain; 51% ML methods used in AML are non-interpretable, 58% studies used sample of old real data; key challenges for researchers are access to recent real transaction data and scarcity of labelled training data; and data being highly imbalanced. Future research directions are, application of XAI techniques to bring-out explainability, graph deep learning using natural language processing (NLP), unsupervised and reinforcement learning to handle lack of labelled data; and joint research programs between research community and industry to benefit from domain knowledge and controlled access to data.

Vehicular traffic plays a significant role in terms of economic development; however, it is also a major source of noise pollution. Therefore, it is highly imperative to model traffic noise, especially for expressways due to their high traffic volume and speed, which produce very-high level of traffic noise. Previous traffic prediction models are mostly based on the regression approach and the artificial neural network (ANN), which often fail to describe the trends of noise. In this paper, a deep neural network-based optimization approach is implemented in two ways: i) using different algorithms for training and activation, and ii) integrating with feature selection methods such as correlation-based feature selection (CFS) and wrapper for feature-subset selection (WFS) methods. These methods are integrated to produce traffic noise maps for different time of the day on weekdays, including morning, afternoon, evening, and night. The novelty of this study is the integration of the feature selection method with the deep neural network for vehicular traffic noise modelling. New Klang Valley Expressway (NKVE) in Malaysia was used as a case study due to its increasing heavy and light vehicles, and the motorbike during peak hours, which result in high traffic noise. The results from the models indicate that the WFS-DNN model has the least mean-absolute-deviation (MAD) of 2.28, and the least root-mean-square-error (RMSE) of 3.97. Also, this model shows the best result compared to the other models such as DNN without integration with feature selection methods, CFS-DNN and the ANN networks (MLP and RBF). MAD improvement of 27% - 47% and RMSE improvement of 25% - 38% was achieved compared to other methods. The study provides a generic approach to key parameter selection and dimension reduction with novel trend descriptor which could be useful for future such modelling applications.

One of the most important tasks in the advanced transportation systems is road extraction. Extracting road region from high-resolution remote sensing imagery is challenging due to complicated background such as buildings, trees shadows, pedestrians and vehicles and rural road networks that have heterogeneous forms with low interclass and high intraclass differences. Recently, deep learning-based techniques have presented a notable enhancement in the image segmentation results, however, most of them still cannot preserve boundary information and obtain high-resolution road segmentation map when processing the remote sensing imagery. In the present study, we introduce a new deep learning-based convolutional network called VNet model to produce a high-resolution road segmentation map. Moreover, a new dual loss function called cross-entropy-dice-loss (CEDL) is defined that synthesize cross-entropy (CE) and dice loss (DL) and consider both local information (CE) and global information (DL) to decrease the class imbalance influence and improve the road extraction results. The proposed VNet+CEDL model is implemented on two various road datasets called Massachusetts and Ottawa datasets. The suggested VNet+CEDL approach achieved an average F1 accuracy of 90.64% for Massachusetts dataset and 92.41% for Ottawa dataset. When compared to other state-of-the-art deep learning-based frameworks like FCN, Segnet and Unet, the proposed approach could improve the results to 1.09%, 2.45% and 0.39%, for Massachusetts dataset and 7.21%, 1.86% and 2.68%, for Ottawa dataset. Also, we compared the proposed method with the state-of-the-art road extraction techniques, and the results proved that the proposed technique outperformed other deep learning-based techniques in road extraction.

Building extraction with high accuracy using semantic segmentation from high-resolution remotely sensed imagery has a wide range of applications like urban planning, updating of geospatial database, and disaster management. However, automatic building extraction with non-noisy segmentation map and obtaining accurate boundary information is a big challenge for most of the popular deep learning methods due to the existence of some barriers like cars, vegetation cover and shadow of trees in the high-resolution remote sensing imagery. Thus, we introduce an end-to-end convolutional neural network called Generative Adversarial Network (GAN) in this study to tackle these issues. In the generative model, we utilized SegNet model with Bi-directional Convolutional LSTM (BConvLSTM) to generate the segmentation map from Massachusetts building dataset containing high-resolution aerial imagery. BConvLSTM combines encoded features (containing of more local information) and decoded features (containing of more semantic information) to improve the performance of the model even with the presence of complex backgrounds and barriers. The adversarial training method enforces long-range spatial label vicinity to tackle with the issue of covering building objects with the existing occlusions such as trees, cars and shadows and achieve high-quality building segmentation outcomes under the complex areas. The quantitative results obtained by the proposed technique with an average F1-score of 96.81% show that the suggested approach could achieve better results through detecting and adjusting the difference between the segmentation model output and the reference map compared to other state-of-the-art approaches such as autoencoder method with 91.36%, SegNet+BConvLSTM with 95.96%, FCN-CRFs with 95.36%% SegNet with 94.77%, and GAN-SCA model with 96.36% accuracy.

This study proposes a new landslide detection technique that is semi-automated and based on a saliency enhancement approach. Unlike most of the landslide detection techniques, the approach presented in this paper is simple yet effective and does not require landslide inventory data for training purposes. It comprises several steps. First, it enhances potential landslide pixels. Then, it removes the image background using slope information derived from a very high-resolution LiDAR-based (light detection and ranging) digital elevation model (DEM). After that, morphological analysis was applied to remove small objects, separate landslide objects from each other, and fill the gaps between large bare soil objects and urban objects. Finally, landslide scars were detected using the Fuzzy C-means (FCM) clustering algorithm. The proposed method was developed based on datasets acquired over the Kinta Valley area in Malaysia and tested on another area with a different environment and topography (i.e., Cameron Highlands). The results showed that the proposed landslide detection technique could detect landslides in the training area with a Prediction Accuracy, Kappa index, and Mean Intersection-Over-Union (mIOU) of 71.12%, 0.81, and 68.52%, respectively. The Prediction Accuracy, Kappa index, and mIOU of the method based on the test dataset were 65.78%, 0.68, and 56.14%, respectively. These results show that the proposed method can be used for landslide inventory mapping and risk assessments.

Multimedia cybersecurity is a prevalent research topic in the digital world due to the rapid progress of digital multimedia and Internet applications. Watermarking, encryption, and steganography schemes are employed to attain multimedia data confidentiality and robustness. However, these schemes are externally applied on trusted computers, and there has been a lack of similar schemes that can be effectively and efficiently enabled through an untrusted transmission medium. In this work, a self-embedding-based High-Efficiency Video Coding (HEVC) transmission and integrity verification framework is presented. This framework is robust and reliable for verifying the integrity of HEVC frames transmitted through insecure communication channels. Firstly, the transmitted HEVC frames are divided into a number of blocks with a certain block size. After that, a discrete transform is used for self-embedding of watermarks from each block into another one depending on a predefined mechanism. The Discrete Wavelet Transform (DWT), Discrete Cosine Transform (DCT), and Discrete Fourier Transform (DFT) are tested for this task. The watermarked HEVC frames are transmitted through a wireless communication channel, and hence they become subject to different attacks and corruptions. At the receiver side, the secret watermarks in each block are sensed with a correlation-based method to discover dubious counterfeit operations. To verify the reliability of the suggested transmission framework and its ability to achieve high protection and robust content verification of the transmitted HEVC frames over insecure communication channels, different HEVC analyses and comparisons are performed. Simulation results demonstrate the suitability of the suggested transmission framework for different multimedia cybersecurity applications. Furthermore, the comparative analysis shows that the DFT is an efficient discrete transform that can be employed with the proposed transmission framework to guarantee a higher HEVC frame integrity. It allows higher sensitivity to simple modifications in the transmitted watermarked HEVC frames. This makes the suggested cybersecurity framework applicable, secure, and appropriate for multimedia integrity verification purposes.

This paper investigates and presents a block-based opto-color cipher using double random phase encoding (DRPE) with different block sizes. The color plainimage is divided into equal-sized blocks and then converted to an optical signal by an optical emitter. The obtained optical signal is encrypted by employing the DRPE technique, which applies two types of phase modulation, time, and Fourier domains. Finally, the optical color cipherimage is, upon detection, converted to digital format by a charge-coupled device digital camera. Experiments and security analysis show that the proposed block-based optical color image cipher using DRPE with increased block size is secure, effective, and including a good immunity to channel noise.