Butterfly segmentation by multi scaled quantum cuts in agro-ecological environment

Abstract

Butterflies have a prominent role in the agro-ecological ecosystems. Some butterfly populations can injure wildlife, vegetation, and even humans in addition to causing harm to flora and fauna. By contrast, the presence of some other ones can help in improving agricultural productivity and preserving the agro-ecological ecosystems. Butterfly segmentation is therefore an initial process that precedes species recognition. In this paper, we propose a new segmentation process that adapts quantum mechanics to be deployed on a multi-layered graph. To achieve a proper butterfly segmentation, we implement efficiently the Schrödinger equation in a propagation process across the different layers of the graph. Furthermore, It is supported by both background and foreground priors guidance coupled to local contrast information. Comparative evaluation suggests that our method has higher resistance than competing methods to artefacts that are inherent to agro-ecological photographs. Our algorithm shows a considerable advantage over single-layered graph based versions when dealing with some image details. It also outperforms some deep learning based methods that achieve high segmentation performance. Unlike these methods, ours does not involve any training step. Thus, it doesn't require high performance equipments or supplementary human labelling operation and does not fall in the problem of generalization.