Environmental microorganism classification using conditional random fields and deep convolutional neural networks

Highlights：

• Application of content-based image analysis to Environmental Microorganism (EM) classification which plays a fundamental role for establishing sustainable ecosystem.

• Building an effective pixel-level feature extractor from scarce training images, by re-purposing a Deep Convolutional Neural Netwrok (DCNN) pre-trained for image classification using large auxiliary data.

• Integration of global features to improve the segmentation quality by providing long-range consistencies among pixel labels​

• Usage of a Conditional Random Field (CRF) to jointly localize and classify EMs by considering the spatial relations among pixel-level features, and their relations to global features.