Detecting and Counting Small Animal Species Using Drone Imagery by Applying Deep Learning

Abstract

This work represents deep learning approach for detecting lizards on the summer grass background. It is the main part of general use case formulation—“how many animals are located now on this substitute habitat. Determine in which parts they prefer to stay”. For this purpose, the U-Net architecture neural network was implemented. Dilated convolution layer was added to usual U-Net. Smoothly blending filter was applied to result probability patches for connecting them in one big probability map without sewed edges. Designed flexible architecture allows to train neural network for pixel-wise semantic segmentation with accuracy value 0.9863 on the tiny dataset.