Our work concentrates on detecting tiny animals in huge air-view images with competitive accuracy and speed. Many environmental organizations investigate distribution of specific species of animals by capturing images from the sky. It is very challenging work for human to check the huge images and mark animals by hand. To check it automatically, we propose the method using CNN-based sliding window. There are many popular works like Faster R-CNN or SSD, that detect multiple objects in image. Despite their state-of-the-art performance, they are not applicable in this situation. Air-view image is huge and animals are tiny as not easy for human to detect. Normal multiple object detection methods are not suitable to detect tiny objects, which are smaller than minimum size threshold. Also, dataset is not suitable to train their networks. The ground-truth dataset doesn’t contain scale information. In this paper, we introduce our own method, from training network using dotted ground truth dataset to detection and classification. Also, we verify the competitive performance of our multi-viewpoint based detection comparing with single viewpoint detection.