The growing adoption of compressed video across diverse applications underscores the demand for efficient action recognition methods. Traditional RGB-based methods face limitations, especially because they depend heavily on computationally intensive optical flow for temporal analysis. We introduce the recursive symmetric temporal interaction transformer (RETIT), which leverages the transformer architecture to enhance global temporal interactions directly from compressed video data. RETIT employs recursive strategies to iteratively refine motion representations and integrates a specialised cross self-attention mechanism to enhance multi-scale spatio-temporal feature extraction. When evaluated on the K400, UCF101, and HMDB51 datasets, RETIT achieved top-1 accuracies of 72.1%, 97.6%, and 75.8%, respectively, surpassing existing state-of-the-art benchmarks. These results highlight RETIT's ability to effectively leverage spatial and temporal modalities, advancing the state of action recognition in compressed video formats.
In order to improve the efficiency and accuracy of topic keyword extraction, a genetic algorithm combining word frequency and inverse document frequency is studied, and a weight evaluation model is constructed using multi-threaded computation and text processing to improve algorithm efficiency. Comparing the performance of different algorithms, it was found that the current research algorithm has an accuracy of 98.9% and a runtime of 0.038 s, while the deep learning method has an accuracy of 97% and a runtime of 0.085 s. Therefore, the research results have demonstrated that the topic word extraction method based on a genetic algorithm for English composition texts not only improves the computational accuracy of topic word extraction but also has good application effects on word frequency extraction in English test questions and provides rich text vocabulary for vocabulary analysis in English test questions and learning platforms, thus having good application potential in the field of English test questions.
This study investigates the robustness of discrete-time adaptive iterative learning control (AILC) under fading channels for nonlinear dynamical systems with multiplicative and additive channel noises. The output and the input fading channel are considered integral components of the system during the analysis process, and thus, the system is reconstructed. Each component of system outputs suffers different multiplicative noise. The framework jointly addresses stochastic multiplicative and additive randomness effects in signal propagation. The variable tracking targets and error dead zones is taken into account in AILC design. The robustness characteristics of the developed AILC methodology are systematically examined. The validity of the developed approach is confirmed through systematic numerical simulations.
This work designs a technique for heart disease prediction by deep learning model. The pre-processing stage is primary phase, and it is considered as most significant process to enhance the technique's performance. The main aim of pre-processing is to transfer raw data into processable data. Moreover, missing data imputation is exploited to carry out pre-processing that is employed to eradicate infinity values for effectual processing. Subsequently, by using the temporal convolutional network, feature fusion is done. Finally, heart disease prediction is done by using ResNet 50 that is trained by proposed serial exponential Siberian tiger optimisation named (SExpSTO) scheme that is derived by combining serial exponential weighted moving average in Siberian tiger optimisation (STO). The performance analysis of proposed algorithm is executed by considering parameters, like accuracy, sensitivity, and specificity. Finally, the experimentation evaluation is performed with maximum accuracy of 95%, maximum sensitivity of 97%, and maximum specificity of 94%.
Fog computing, characterised by its dynamic and heterogeneous nature, necessitates efficient resource provisioning to maximise resource utilisation. Existing methods often suffer from computational overhead and scalability limitations. This research proposes a novel resource provisioning approach for fog environments to address these challenges. The user request is first transformed into a four queue model, in which very urgent and urgent requests are placed in the first and second priority queues to offload to the closest available fog resource while considering deadliness and latency. Here, the selection of a suitable fog node selection for flawless workload execution, an improved coati optimisation algorithm (ICOA) is developed. The proposed method is evaluated in terms of packet delivery ratio (PDR), processing time, end to end delay, throughput, energy consumption, and cost and acquired the value of 97.5 (%), 105.061 (ms), 44.807 (ms), 95.1826 (%), 21.2539 (Kwh), and 0.42402 ($) respectively.
Many-objective optimisation challenges traditional Pareto-based evolutionary algorithms. This work introduces a self-adjusting dominance relation using an adaptive projection plane, enhancing convergence by leveraging distances to the plane and between projection points without extra parameters. It also proposes a PBI-function-based method to balance convergence and diversity in solution screening. These techniques are integrated into the NSGA-II framework, creating the MaOEA/APP-PBI algorithm. Evaluated on 5-, 10-, 15-, and 20-objective DTLZ and WFG problems using IGD and HV metrics, MaOEA/APP-PBI outperforms six leading algorithms. Results demonstrate its significantly superior convergence and diversity across various objectives, highlighting its effectiveness for many-objective optimisation.
IoT-based attacks have steadily increased in quantity due to the growing use of IoT devices. The method is useful for resolving optimisation issues since it finds a balance between exploitation and exploration. However, the current intrusion detection systems may have trouble spotting intricate attack patterns if they are not familiar with IoT gadgets and their vulnerabilities to mitigate this challenge. This work introduces the IoT botnet attack detection of customised pelican optimisation algorithm (IoT-BADS-CPOA). Initially, the data normalisation is carried out. From the normalised data, higher-order features, improved correlation, statistical features, and improved technical features are derived as the features. The features are extracted; the presence of attacks is detected by a deep ensemble of classification models. Particularly, a new self-improved version of pelican algorithm is introduced, to tune DQN. In particular, the C-POA has an accuracy of 93.86%.
