Scene-Based Event Detection using Grey Wolf Optimization Enabled BiLSTM Classification

Event detection is a crucial task that enables analyzing the rapid moment of events for rectifying the issues associated with the event that occurred. Despite, several traditional methods have been developed for event detection, the methods lag in identifying the effective localization of key moments and the classification of localized key moments. To address the aforementioned issues and to differentiate the events that contain similar objects or scenes, a novel framework is developed utilizing the Bi-directional Long Short Term Memory with Grey Wolf Optimization (BiLSTM-GWO) for efficient feature extraction and classification. The novel BiLSTM with GWO framework in which the features including the Scale Invariant Feature Transform(SIFT), Hybrid ternary pattern involving the local directional pattern(LDP), local binary pattern (LBP), and local optimal-oriented pattern (LOOP), and Resnet 101 are extracted to assist in learning the relationship between the scenes. For effective event detection, the BiLSTM utilizing the long-term dependency is used and the importance of the developed model relies on the GWO utilizing the encircling and hunting strategies that boost the performance of the classifier. The efficiency of the BiLSTM with GWO approach is evaluated as 91.155%, 91.7%, and 91.505% for sensitivity, specificity, and accuracy respectively with other traditional techniques concerning the USED dataset.