Enhanced Video Summarization Using BiLSTM Encoder-Decoder with Dual Attention and Particle Swarm Optimization

We need efficient techniques of summarizing that keep important information while decreasing redundancy because video data is growing at an exponential rate. Using a Bidirectional Long Short-Term Memory (BiLSTM) encoder-decoder model, an improved attention mechanism, and Particle Swarm Optimization (PSO), this research introduces a novel framework for video summarization. In order to provide a comprehensive contextual knowledge of video frames, the BiLSTM is used to record forward and backward temporal dependencies. Using additive and multiplicative dual attention algorithms to dynamically emphasize crucial frames based on their temporal relevance significantly improves the decoder's performance. In order to guarantee the production of brief and accurate video summaries, PSO enhances the process of frame selection by increasing their relevance scores. Extensive tests on state-of-the-art F1 datasets, such as SumMe and TVSum, validate the efficacy of the suggested method. Thanks to its adaptability and scalability, the framework shows promise as a solution for various video summary tasks, which in turn helps to improve efficient video data management.