All Times ET
Keywords: Convolutional neural networks, classification, pollen, low dimensional data
Pollen is one of the most significant allergens, and there is a necessity for automatic real-time detection. Rapid-E particle detector produces three low dimensional but complex data types, representing morphological and chemical pollen properties. This study implements vanilla CNNs, residual, and inception networks to Rapid-E data. We show that residual networks are superior to vanilla and inception networks in the low dimensional settings. Furthermore, we demonstrate improved performance with residual networks when pretrained convolutional branches are implemented in multimodal training.