Today, public areas, such as airports, hospitals, city centers are monitored by surveillance systems. The widespread use of surveillance systems reduces security concerns while creating an amount of video data that cannot be examined by people in real-time. Therefore, the concept of automatic understanding of video activities has raised the standards of security camera systems. In this paper, we propose a framework (OF-ConvAE-LSTM) to detect anomalies using Convolutional Autoencoder and Convolutional Long Short-Term Memory in an unsupervised manner. Besides the deep learning model, the feature extraction stage based on dense optical flow is applied in the framework to obtain the velocity and direction information of foreground objects. The experiments were carried out on three popular public datasets consisting of Avenue, UCSD Ped1, and UCSD Peds2. The experimental results have shown that the proposed framework models the complex distribution of the pattern of regular motion changes with high accuracy. Besides, this method was observed to outperform state-of-the-art approaches based on unsupervised and semi-supervised deep learning models.

Accurate and efficient localization of the optic disk (OD) in retinal images is an essential process for the diagnosis of retinal diseases, such as diabetic retinopathy, papilledema, and glaucoma, in automatic retinal analysis systems. This paper presents an effective and robust framework for automatic detection of the OD. The framework begins with the process of elimination of the pixels below the average brightness level of the retinal images. Next, a method based on the modified robust rank order was used for edge detection. Finally, the circular Hough transform (CHT) was performed on the obtained retinal images for OD localization. Three public datasets were used to evaluate the performance of the proposed method. The optic disks were successfully located with the success rates of 100%, 96.92%, and 98.88% for the DRIVE, DIARETDB0, and DIARETDB1 datasets, respectively.

Industrial automation and control systems have gained increasing attention in the literature recently. Their integration with various systems has triggered considerable developments in critical infrastructure systems. With different network structures, these systems need to communicate with each other, work in an integrated manner, be controlled, and intervene effectively when necessary. Supervision Control and Data Acquisition (SCADA) systems are mostly utilized to achieve these aims. SCADA systems, which control and monitor the connected systems, have been the target of cyber attackers. These systems are subject to cyberattacks due to the openness to external networks, remote controllability, and SCADA-architecture-specific cyber vulnerabilities. Protecting SCADA systems on critical infrastructure systems against cyberattacks is an important issue that concerns governments in many aspects such as economics, politics, transport, communication, health, security, and reliability. In this study, we physically demonstrated a scaled-down version of a real water plant via a Testbed environment created including a SCADA system. In order to disrupt the functioning of the SCADA system in this environment, five attack scenarios were designed by performing various DDoS attacks, i.e., TCP, UDP, SYN, spoofing IP, and ICMP Flooding. Additionally, we evaluated a scenario with the baseline behavior of the SCADA system that contains no attack. During the implementation of the scenarios, the SCADA system network was monitored, and network data flow was collected and recorded. CNN models, LSTM models, hybrid deep learning models that amalgamate CNN and LSTM, and traditional machine learning models were applied to the obtained data. The test results of various DDoS attacks demonstrated that the hybrid model and the decision tree model are the most suitable for such environments, reaching the highest test accuracy of 95% and 99%, respectively. Moreover, we tested the hybrid model on a dataset that is used commonly in the literature which resulted in 98% accuracy. Thus, it is suggested that the security of the SCADA system can be effectively improved, and we demonstrated that the proposed models have a potential to work in harmony on real field systems.

Sentimental Analysis is considered a computational strategy that helps in identifying and assessing the emotions of people via text documents. Tools and different methods have been adopted for determining both positive and negative emotions in the form of text data analytics by using Machine and Deep Learning techniques. Experimentally, it has been shown that the accuracy of existing text classification models such as Bi-LSTM, Decision Tree, and Ensemble Classifiers is limited by poor quality data, inappropriate hyperparameter tuning, and model-specific bias levels. Additionally, these models are prone to overfitting, high computational overhead, and longer training time. To overcome these limitations, we proposed a hybrid binary classification framework by combining Deep sequential features with the Random Forest (RF) technique. The approach is implemented in four phases: Initially, data preprocessing is performed by employing a Vader sentiment package. In the second step, the deep Long Short Term Memory (LSTM) model was employed to extract deep sequential features corresponding to sad and happy emotions. In the third phase, a bi-orthogonalization algorithm with principal component Analysis (PCA) and Singular Value Decomposition (SVD) was employed to minimize the redundancy and maximize the relevance of extracted features. Finally, a five-fold cross-validation technique was implemented to discriminate sad and happy emotions using the Random Forest (RF) algorithm. Eventually, a grid search approach was implemented for hyperparameter tuning and results were compared with five baseline algorithms (Vanilla LSTM (VLSTM), Support Vector Machine (SVM), Gradient Boosting Machine (GBM), Naïve Bayes (NB), Ada Boost Algorithm (ABA). The experimental outcomes revealed that the proposed model achieved an accuracy rate of 99.631% on the 4000 stories dataset which was superior to all five state-of-the-art methods with a margin of 4.63%, 10.7%, 19.44%, 21%, and 56.5%, respectively. Interestingly, the proposed model realized improved results in terms of other conventional performance metrics also such as precision, recall, specificity, and time complexity. Overall, the proposed model has great potential in educational institutions, child psychology research, and child-friendly content moderation, generally helping in the understanding of the emotions and experiences of children in the digital realm.