PalmNeXt: a ConvNeXt-based deep learning model for pest detection in date palm leaves

TL;DR

This paper introduces a lightweight deep learning model for detecting pests in date palm leaves, achieving high accuracy and outperforming existing methods.

Contribution

The novel contribution is a preprocessing-augmented ConvNeXt-Tiny framework tailored for pest detection in small and variable agricultural datasets.

Findings

The proposed model outperformed custom and state-of-the-art baselines in accuracy, precision, recall, and F1-score.

The tailored preprocessing pipeline significantly improved feature quality and model performance.

The model provides a scalable and lightweight solution for precision agriculture pest detection.

Abstract

Automated pest detection is essential for timely and accurate crop monitoring, yet many existing approaches rely on manual inspection or computationally heavy models that struggle with small and variable datasets. To address these challenges, we introduce an enhanced ConvNeXt-Tiny–based framework that incorporates a tailored preprocessing pipeline to improve feature quality and overall performance. The model is evaluated on an RGB image dataset of 3,000 date palm leaf samples across four classes (Bug, Dubas, Healthy, Honey). Its performance is compared against two custom baselines, CNN-Attention and ResNet13-Attention, as well as state-of-the-art models including ViT, ECA-Net, and the standard ConvNeXt-Tiny. Experimental results show that our preprocessing-augmented ConvNeXt-Tiny achieves the highest accuracy, precision, recall, and F1-score, outperforming both custom and state-of-the-art baselines. These findings demonstrate the effectiveness of the proposed lightweight solution for scalable and high-accuracy pest detection in precision agriculture.

Figures