Learning of Counting Crowded Birds of Various Scales via Novel Density Activation Maps

Abstract

The previous counting methods trained by the density map regression scheme fail to precisely count the number of birds in crowded bird images of various scales. This is due to the coarseness of the manually created target density maps. In this paper, we propose a new counting scheme, called DAM counting, which generates our-first-proposed density activation map (DAM). DAM is a CNN perspective density map that has high activation values where the network focuses on for precise counting of birds. The network is trained to autonomously learn where and how much the DAM should be activated so that the sum of all values in the DAM estimates the number of birds. Moreover, our DAM counting scheme incorporates two segmentation regularizers that enable precise counting of birds with various scales and appearance. Our DAM counting scheme can effectively substitute the existing density map regression scheme, bringing in a remarkable increase in the counting accuracy. We also propose the first crowded bird dataset, called CBD-6000, which is very valuable for crowded bird counting research.

A general deep learning model for bird detection in high‐resolution airborne imagery

“…Birds are a major indicator of ecosystem health and represent an important intersection of remote sensing and conservation (Gregory & van Strien, 2010). Airborne monitoring of birds using unoccupied aerial vehicles (UAVs) and airplanes is increasingly common owing to the need to monitor birds on large scales (Dulava et al, 2015; Groom et al, 2013; Kim & Kim, 2020; Pfeifer et al, 2021). Much of this work involves hand counting birds in imagery using human annotators.…”

Automating sandhill crane counts from nocturnal thermal aerial imagery using deep learning

“…Hong et al (2019) experimented with several different DL object detection models for bird species and found that Faster R-CNN (Ren et al, 2015) consistently performed well. To address situations with very dense aggregations of bird targets, several studies have proposed novel DL-based counting architectures (Arteta et al, 2016;Kellenberger et al, 2021;Kim & Kim, 2020). In particular, Kim and Kim (2020) integrated density estimation into their architecture, with strong counting performance on images of crowded birds from a variety of perspectives.…”

“…To address situations with very dense aggregations of bird targets, several studies have proposed novel DL‐based counting architectures (Arteta et al., 2016; Kellenberger et al., 2021; Kim & Kim, 2020). In particular, Kim and Kim (2020) integrated density estimation into their architecture, with strong counting performance on images of crowded birds from a variety of perspectives. Even with these advancements, factors that affect model performance, such as spatial resolution, are generally unexplored.…”

Identifying Habitat Elements from Bird Images Using Deep Convolutional Neural Networks

“…This remarkable advantage makes researchers only need to collect corresponding data according to the target when using this technology, instead of making various attempts for feature selection and extraction. To assist ecologists and zoologists in rapidly and effectively processing large-scale bird image data, computer vision research has long dealt with bird image analysis-related problems, such as bird detection [ 24 ], the counting of crowded birds [ 25 , 26 ], fine-grained classification [ 27 , 28 , 29 , 30 ] of birds, and even individual recognition with small birds [ 31 ], using DCNNs. DCNNs have achieved surprising results in these tasks.…”