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

TL;DR

This paper presents FCBNet, a lightweight, parameter-efficient weed segmentation model using a frozen ConvNeXt backbone and feature correction blocks, outperforming existing models in accuracy and efficiency on multispectral aerial imagery datasets.

Contribution

Introduction of FCBNet, a novel weed detection model with a frozen backbone and efficient convolutions, significantly reducing parameters and training time while improving accuracy.

Findings

FCBNet achieves over 85% mIoU on multiple datasets.

The model reduces trainable parameters by more than 90%.

Training time is significantly lower compared to other models.

Abstract

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.