| Summary: | Automated detection of intestinal parasites is crucial for improving diagnostic efficiency and accuracy in parasitology. This study evaluates the performance of three object detection models: Faster RCNN with ResNet back-bone, Faster RCNN with RetinaNet backbone, and YOLOv8. A dataset comprising 2000 microscopic images of two parasite species, Ascaris lumbricoides and Trichuris trichiura, was used. The dataset was split into 1500 images for training, 300 for validation, and 200 for testing. Experimental results show that Faster RCNN with RetinaNet achieved the highest Average Precision (AP) across varying Intersection over Union (IoU) thresholds, demonstrating its robustness. YOLOv8 exhibited superior precision at low confidence thresholds, while Faster RCNN with ResNet demonstrated strong recall consistency. These findings provide a comparative analysis, highlighting the strengths and limitations of each model for reliable and efficient intestinal parasite detection, suggesting the integration of ensemble models to combine RetinaNet’s robustness and YOLOv8’s precision-recall capabilities for optimized results.
|
|---|