Adaptive Landmine Detection and Recognition in Complex Environments using YOLOv8 Architectures

Authors : Ahmed Shahab Ahmed Alslemani, Govar Abubakr

Pages : 110-120

Doi:10.58769/joinssr.1542886

View : 29 | Download : 32

Publication Date : 2024-12-20

Article Type : Research Paper

Abstract :Landmine detection and recognition represent critical tasks in humanitarian and military operations, aiming to mitigate the devastating impact of landmines on civilian populations and military personnel. Landmine detection and identification using computer vision offers several advantages. Safety is enhanced, given the reduced exposure to humans in dangerous environments. Advanced algorithms are applied to increase the performance of a computer system operating with high accuracy and efficiency in the location of hidden. Fast detection is made possible by real-time processing, which is essential for time-sensitive processes. Furthermore, unlike human operators, computer vision can work continuously without getting tired. The efficacy of these systems is further enhanced by their capacity to adapt to various environments. This abstract explores the application of You Only Look Once (YOLO), a state-of-the-art object detection algorithm, in the domain of landmine detection and recognition. YOLO offers real-time performance and high accuracy in identifying objects within images and video streams, making it a promising candidate for automating landmine detection processes. By training YOLO on annotated datasets containing diverse landmine types, terrains, and environmental conditions, the algorithm can learn to detect and classify landmines with remarkable precision. Integrating YOLO with unmanned aerial vehicles (UAVs) or ground-based robotic systems enables rapid and systematic surveying of large areas, enhancing the efficiency and safety of demining operations. The YOLOv8 is employed in this research to address the issue of missed detection and low accuracy in real-world landmine detection. For this study, we have assembled a data set of 1055 photos that were shot in various lighting and backdrop situations. In the experiment employing picture data, we obtained very good results with mAP = 93.2%, precision = 92.9%, and recall = 84.3% after training the model on the dataset numerous times. According to experimental results, the YOLOv8 has better detection accuracy and recall based on the landmine dataset.
Keywords : Kara mayını, YOLO, Algılama ve Tanıma, Bilgisayarlı Görüntüleme