SegFormer-based nectar source segmentation in remote sensing imagery

TL;DR

This paper introduces an improved SegFormer model for identifying nectar-producing plants in satellite images, enhancing beekeeping and ecological monitoring.

Contribution

The novel contribution is integrating CBAM, residual structures, and spatial enhancement to improve SegFormer for nectar source segmentation.

Findings

The improved model achieved 91.05% mIoU, outperforming the baseline SegFormer.

Mean pixel accuracy improved to 95.02%, with precision and recall reaching 95.40% and 95.02%.

The method supports real-time precision beekeeping and ecological monitoring.

Abstract

Beekeepers often face challenges in accurately determining the spatial distribution of nectar-producing plants, which is crucial for informed decision-making and efficient beekeeping.

In this study, we present an efficient approach for automatically identifying nectar-producing plants using remote sensing imagery. High-resolution satellite images were collected and preprocessed, and an improved segmentation model based on the SegFormer architecture was developed. The model integrates the CBAM attention mechanism, deep residual structures, and a spatial feature enhancement module to improve segmentation accuracy.

Experimental results on rapeseed flower images from Wuyuan County demonstrate that the improved model outperforms the baseline SegFormer model. The mean Intersection over Union (mIoU) increased from 89.31% to 91.05%, mean Pixel Accuracy (mPA) improved from 94.15% to 95.02%, and both mean Precision and mean Recall reached 95.40% and 95.02%, respectively.

The proposed method significantly enhances the efficiency and accuracy of nectar plant identification, providing real-time and reliable technical support for precision beekeeping management, smart agriculture, and ecological monitoring. It plays a key role in optimizing bee colony migration, improving collection efficiency, and regulating honey quality.

Figures