Bird small target detection method based on density map estimation

Abstract

The invention discloses a bird small target detection method based on density map estimation, and relates to the ecological monitoring technology based on computer vision. The bird small target detection method comprises a training stage and a detection stage, including learning a density map by utilizing a method based on instance learning, estimating the position of a bird small target on the density map through a sliding window and a method for solving a local extreme value in the window, carrying out super-pixel-based bounding box estimation, and realizing bird small target detection under a small sample training set. The bird small target detection method reduces the dependence on the scale of the training set while reducing the manual marking work of a large number of tedious training samples, can preferably match the shape and the contour of a small target, and is high in the detection precision.

Description

technical field [0001] The invention belongs to the field of computer vision technology and target detection technology, and relates to an ecological monitoring technology based on computer vision, in particular to a small bird target detection method based on density map estimation. Background technique [0002] Birds are one of the most studied and frequently investigated animal groups due to their wide distribution, diverse species, and easy identification. They are also a group of animals that are extremely sensitive to habitat changes and environmental changes. The composition, quantity, diversity and community characteristics of bird species can directly reflect the suitability of habitats, the health of ecosystems and biodiversity, the degree of disturbance of human activities on ecosystems, and the impact of land use and landscape changes on ecosystems. degree, and the quality of the regional ecological environment. So far, the field monitoring stations mainly follo...

AI Technical Summary

Problems solved by technology

The small scale of bird individuals, overlapping occlusions and other issues lead to the loss of effective feature information in the image; at the same time, a large number of densely distributed small bird targets bring difficulties to manually label training samples

In view of the above problems, the detection effect of these existing mainstream target detection algorithms in the small bird target detection task has declined, and the detection effect is not good.

Images