Nannofossil Image Enhancement for Improved Detection Using Modified Contrast Stretching Method and YOLO-v8

This research focuses on enhancing the quality of nannofossil images, specifically the discoaster and coccolith genus, to improve the performance of detection and prediction models using advanced deep learning techniques. The study employs Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN) for image resolution enhancement and a modified mean-standard deviation contrast stretching method for contrast enhancement. The You Only Look Once version 8 (YOLOv8) algorithm is then used to detect enhanced images. Evaluation metrics, including precision and recall, were used to assess the impact of image enhancement on detection performance. Results indicate that the modified mean-standard deviation contrast stretching method significantly improves image clarity and object detectability, achieving a precision of 0.592 and a recall of 0.211. However, these results are still lacking to be considered as good result for object detection. Thus, further improvement is required to analyze the data. This research is expected to make significant contributions to the study of paleoclimatology and the development of deep learning-based object detection methods in the future.