Improved Salp Swarm Optimization with Deep Transfer Learning based Visual Places Recognition

Visual place recognition (VPR) includes inferring a geographical location of a single image with wide-ranging applications in consumer photography, robotics, archival repositories, and social media. The representation of ideal image for VPR maps each image that captures a similar location closer to one another, nevertheless of different illuminations, viewpoints, capturing sensors, and appearances. It has received considerable attention in various namely robotics and computer vision. In recent times, researcher workers have applied state-of-the-art deep learning (DL) techniques to address these problems. However, a growing number of studies have projected an innovative place recognition methodology dependent upon DL, new automated models become essential. This study develops an Improved Salp Swarm Optimization with Deep Transfer Learning based Visual Places Recognition (ISSODTL-VPR) approach. The aim of the ISSODTL-VPR model is to recognize the visual places accurately, automatically, and timely. In the presented ISSODTL-VPR model, the initial stage of image preprocessing using bilateral filter (BF) is applied. Followed by, EfficientNet with Squeeze-excitation block is utilized for feature extraction purposes. Then, the ISSO algorithm is used as a hyperparameter optimizer of the EfficientNet model. At last, Manhattan distance metric is applied for the recognition of places in the presented ISSODTL-VPR model. The experimental validation of the ISSODTL-VPR technique was tested and the results are examined under many measures. The experimental outcomes demonstrate the improvement of the ISSODTL-VPR technique over other existing techniques.