Enhancing Object Detection in Autonomous Navigation with LSTM-YOLOv8 and Hybrid Bio-Inspired Optimization

Object detection is a key challenge in Autonomous Navigation. To enhance the object detection accuracy, this research work proposes an innovative hybrid framework that integrates the strengths and exceptional qualities of Long Short-Term Memory (LSTM) and YOLOv8, along with bio-inspired optimization algorithms. We present a synergistic framework to optimise the deep learning (DL) models' hyperparameters by utilising the Humboldt Squid Optimisation Algorithm (HSOA) and the Polar Coordinate Bald Eagle Search Optimisation Algorithm (PCBESOA). The experimental results surpass current methods in object detection accuracy, used primarily in Autonomous vehicles, demonstrating significant improvements across multiple datasets. The suggested method exhibits potential in resolving issues with scale fluctuation, complex backgrounds, object occlusions and detection of static as well as dynamic objects. The results and findings reveal that the proposed hybrid methodology significantly surpasses existing object detection techniques across various datasets, achieving an enhancement of up to 3.3% in mAP@[0.5:0.95] on the COCO dataset.