Parasitic Egg Detection and Classification by Utilizing the YOLO Algorithm with Deep Latent Space Image Restoration and GrabCut Augmentation

Yohanssen Pratama, Yuki Fujimura, Takuya Funatomi, Yasuhiro Mukaigawa

Abstract: A parasitic egg could cause infections and become significant diseases, especially in countries with poor sanitation and hygiene. Its detection which is manually conducted by humans using a microscope could be time-consuming and potent to misclassify. The goal of this research is to get a method that has a good accuracy to automatically detect and identify the parasitic egg. Also, we have a challenge here because the dataset has different resolutions, lighting, and setting conditions. Here we try to utilize the YOLO algorithm by doing image augmentation and restoration to the dataset, we use deep latent space translation for restoration and a simple GrabCut algorithm to generate an augmented image. Furthermore, we apply hyperparameter tuning in image resolution and see how we could utilize the YOLO algorithm to do fast and accurate classification. From the result, we get the combination could give better results in classifying the parasitic egg which the mIoU results reached 71.7 percent.