Figure 1. Results of the Layers method: this permits a precise count of the birds incubating in a breeding colony, identification of the species, and activity, i.e., whether the bird is incubating or standing on the ground. Red circles - Black-headed Gulls incubating, black circles - Black-headed Gulls standing on the ground or in the water, yellow circles - Greylag Geese incubating and nest with eggs, blue circles – incubating Common Terns, white squares - birds with an unidentified activity, possibly dead.
A separate comparison of 43 images was used for the results obtained with the predictive machine learning model. The machine learning method requires additional assumptions and more preparatory work. 1) Images must be 8 bits. 2) The bird size needs to be roughly the same in all images, so downsizing is necessary using the resizing factor calculated from the image with the objects of the fewest pixels. 3) The images must be pseudo-normalized via automatic adjust brightness-contrast in Fiji software (Schindelin et al. 2012, Buchholz et al. 2020).
Passin Bablok Regression was used to compare the methods. This is a linear regression procedure with no special assumptions regarding sample distribution or measurement errors. The result does not depend on the assignment of the methods or instruments. The slope and intercept are calculated with their 95% confidence interval. These confidence intervals are used to determine whether there is only a chance difference between the slope and 1 and between the intercept and 0 (Passing & Bablok 1983, Bilić-Zulle 2011). The statistical analyses were carried out using the software program R (R Core Team. 2021).
If the methods were comparable (not significantly different) with the proxy method, the analysis execution time was compared in the next step.
Results