The application of convolutional neural network for pollen bearing bee classification

Abstract

The article presents the classification of images with pollen bearing bees using convolutional neural network (CNN). The aim is to find out a sufficient configuration of CNN required for future implementation on low-cost FPGA. A new dataset with bee images was collected on the entrances to several beehives. hidden layers with up to 15 15 x 15 down to 3x3 filter sizes. The CNN configured to three hidden layers 7–7, 5–5, 3–3 was selected for future application as a trade off between accuracy 94% and number of required arithmetic operations.