How the Quality of AI-Generated Microscopic Images Affects Image Classification Accuracy?

What is it about?

Creating high-quality images for training AI models in identifying plankton, especially cyanobacteria, is challenging and requires biologists' expertise. To improve model performance, data augmentation techniques, including those using Generative Adversarial Networks (GANs), can be helpful. However, GANs usually need large datasets to produce good results. To address this, we used the StyleGAN2-ADA model with a dataset of 9 cyanobacteria genera and non-cyanobacterial microalgae. We evaluated the quality of the generated images through both expert reviews by biologists and quantitative metrics. The experts checked for any issues in shape, texture, and color that could hinder visual classification, while quantitative methods assessed image quality based on perceptual features. The high-quality images that passed these evaluations will be used to train classification models, and we'll compare the performance improvement over the original dataset. This thorough assessment ensures that the generated images are useful for enhancing the AI models' accuracy.

Featured Image

Photo by Tayfun Dikmen on Unsplash

Photo by Tayfun Dikmen on Unsplash

Why is it important?

Using the StyleGAN2-ADA technique to create additional images for the cyanobacteria dataset has improved the performance of classification tasks. This means that the new images are of good quality and don't negatively affect the dataset, helping to make the classification process more accurate.