Ecological monitoring programs play a crucial role in understanding population dynamics and are fundamental to conservation planning. These programs enable researchers to describe natural patterns, detect disturbances, and provide essential information for effective management and decision-making.
New technologies have revolutionized data collection in natural systems, enhancing the precision and accuracy of monitoring programs. In this sense, drones have gained widespread use due to their accessibility, cost-effectiveness, versatility, flight autonomy, and data-collection capabilities. They have found applications in monitoring processes across fields like agriculture, forestry, and ecology, offering advantages such as replicable flight paths and reduced sampling effort. While drones enhance spatial data accuracy, they can also increase data collection volume and analysis time, posing research challenges.
To address this issue, image processing and analysis automation has emerged as a promising research area. Automation streamlines the analysis of photographs and videos, reduces observer bias, and enables standardization and replicability. Automatic recognition methods rely on spectral properties, pattern recognition (e.g., shape and texture), and filters to enhance contrast between the object of interest and the background. These methods facilitate systematic monitoring of multiple species, reduce analysis time, and contribute to the early detection of wildlife behavioral or population changes.
A.- A group of swans in the water, B.- pairs of swans in the vegetation, C.- family groups, adult swans and young swans, D.- pairs of swans in nests
This study uses drone imagery and supervised classification methods to present an automatic counting protocol for black-necked swans (Cygnus melancoryphus) under natural conditions. The study area was the Carlos Anwandter Sanctuary, a coastal wetland in Chile with additional sites along the Cruces River. The researchers conducted 110 survey missions between July 2017 and October 2018 using a drone equipped with an HD standard camera recorder (DJI Phantom 3 advanced drone). The automatic recognition system comprises two steps: (i) The spectral signature is based on the range of spectral values for each pixel in each band (red, green, and blue) for selected individuals. (ii) The shape attributes encompass various measures of the object’s size, perimeter, area, and shape index. The automatic recognition system showed promising results, with an overall accuracy of approximately 96.74%.
However, the system had some limitations and sources of error, including difficulty distinguishing extremely close individuals, recognizing young swans, issues with birds in vegetation, and identifying flying swans or distorted images. The accuracy of the recognition system varied depending on the brightness. In images without brightness, the system performed well, with an accuracy of approximately 98.18%. However, the system’s accuracy dropped significantly to about 82.20% in images with high brightness. However, we reduced the brightness effect by incorporating an ND 8 Polarized filter in the camera lens. The study highlights the effectiveness of using drones and automated recognition systems for wildlife monitoring, specifically for counting black-necked swans. The automated system showed high accuracy in identifying swans in images without brightness, demonstrating its potential as a valuable tool for ecological monitoring. Automated recognition systems have the potential to enhance the efficiency and accuracy of ecological monitoring efforts, but researchers must consider the specific challenges and limitations associated with different environmental conditions and species.
The full article can be found here