A Parameter-efficient Convolutional Approach for Weed Detection in Multispectral Aerial Imagery

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ramos, Leo Thomas, Sappa, Angel D.
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908870750765056
author Ramos, Leo Thomas
Sappa, Angel D.
author_facet Ramos, Leo Thomas
Sappa, Angel D.
contents We introduce FCBNet, an efficient model designed for weed segmentation. The architecture is based on a fully frozen ConvNeXt backbone, the proposed Feature Correction Block (FCB), which leverages efficient convolutions for feature refinement, and a lightweight decoder. FCBNet is evaluated on the WeedBananaCOD and WeedMap datasets under both RGB and multispectral modalities, showing that FCBNet outperforms models such as U-Net, DeepLabV3+, SK-U-Net, SegFormer, and WeedSense in terms of mIoU, exceeding 85%, while also achieving superior computational efficiency, requiring only 0.06 to 0.2 hours for training. Furthermore, the frozen backbone strategy reduces the number of trainable parameters by more than 90%, significantly lowering memory requirements.
format Preprint
id arxiv_https___arxiv_org_abs_2603_06655
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle A Parameter-efficient Convolutional Approach for Weed Detection in Multispectral Aerial Imagery
Ramos, Leo Thomas
Sappa, Angel D.
Computer Vision and Pattern Recognition
Artificial Intelligence
We introduce FCBNet, an efficient model designed for weed segmentation. The architecture is based on a fully frozen ConvNeXt backbone, the proposed Feature Correction Block (FCB), which leverages efficient convolutions for feature refinement, and a lightweight decoder. FCBNet is evaluated on the WeedBananaCOD and WeedMap datasets under both RGB and multispectral modalities, showing that FCBNet outperforms models such as U-Net, DeepLabV3+, SK-U-Net, SegFormer, and WeedSense in terms of mIoU, exceeding 85%, while also achieving superior computational efficiency, requiring only 0.06 to 0.2 hours for training. Furthermore, the frozen backbone strategy reduces the number of trainable parameters by more than 90%, significantly lowering memory requirements.
title A Parameter-efficient Convolutional Approach for Weed Detection in Multispectral Aerial Imagery
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2603.06655