Investigating the Efficiency of Deep Learning Models in Bioinspired Object Detection
Sunday Adeola Ajagbe, Olukayode Oki, Matthew Abiola Oladipupo, Andrew Chinonso Nwanakwaugwu
Abstract
Object detection is the process of using a camera to track an object or a group of objects over time. It has numerous applications like human-computer interactions (HCI), security and surveillance, bioinspired approach, traffic control, and public areas such as airports, subway stations, and event centres. This application has prompted an extensive study in the field of computer vision for more than the last decade now. Visual recognition which includes picture categorization, localization, and detection, is at the heart of all of these applications and has gathered a lot of research attention. These visual identification algorithms have achieved extraordinary performance thanks to considerable advancements in neural networks, particularly deep learning (DL). Despite all successes recorded in object detection through the use of DL models, the experimental-based approach to investigate the performance of DL models for bioinspired object detection (BOD) still remain an open issue. Thus, this paper investigates the efficiency of DL models in BOD using six (6) performance metrics. Based on the literature, eight common DL models were selected for the experiment. Beetles Bee and Morder hornet were contained in the bioinspired datasets that were used as object detection images on MATLAB 2018a. The results show that CNN outperformed the other 7 DL models in the achieved training time, detection accuracy, sensitivity, specificity, and precision metrics. These results suggest CNN as the efficient model that can be considered taking into account the focus of the project at hand. The modification of the eight DL models' layers and architectures to study their efficiency under different scenarios was highlighted as the future scope of this study.