Answers are grounded in this paper and cite their sources.
Within the realm of agricultural production, the fruit and vegetable industry always holds a crucial position. However, the rampant spread of fruit fly pests has severely hindered the healthy development of this industry. Traditional control methods mostly rely on chemical pesticides, which not only cause environmental pollution and damage but also pose potential risks to human health. Therefore, developing precise and efficient fruit fly segmentation technologies has become a core demand in agricultural research and production practice. Currently, the existing target segmentation methods often ignore the spatial relationships between segmentation targets when dealing with unstructured data, resulting in decreased segmentation accuracy. Moreover, supervised and semi-supervised learning methods require specific labeled data, leading to poor segmentation performance and high time complexity. This paper proposes a weakly supervised mobile target segmentation framework based on difference extraction and multi-dimensional information fusion (DEMFF-WMBTS). This framework converts the MOS problem into a weakly supervised node classification task through the graph convolutional network (GCN), combines multi-dimensional feature fusion technology to capture the spatial relationships of targets, and significantly improves the segmentation accuracy. Additionally, this paper proposes a difference extraction module, which aims to use an intelligent feature selection mechanism to judge redundant information and solve the problem of interference caused by background changes and target movement on the results. Experiments on the self-made Drosophila Segmentation Dataset (DSD) and the GBIF public dataset have fully verified the effectiveness of DEMFF-WMBTS, with an F1 value improvement to 0.771.
No takes yet. Share an insight, caveat, or question.
Deng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: