Improved Prediction of Geckler Classification in Gram Stained Smears Images for Sputum

In this paper, we predict Geckler classification as Geckler classes and quality classes in Gram stained smears images for sputum, by using image classification and object detection. Here, we adopt VGG, MobileNet, DenseNet, RegNet, ConvNeXt, ViT and EfficientNet as image classifiers and YOLO11, HIC-YOLO11 and SOD-YOLO11 as object detectors. Note that, whereas the image classifiers classify the classes of images directly, the object detectors first detect buccal squamous epithelial (BSE) cells and leukocytes in the image and then predict the class of the image by the number of them.