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

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Automated pest detection is crucial for efficient crop monitoring.
• Existing methods often require manual inspection or are computationally intensive, limiting scalability.
• Challenges include handling small, variable datasets common in agricultural applications.

Purpose of the Study:

• To develop a high-accuracy, computationally efficient automated pest detection framework for date palm crops.
• To enhance the performance of the ConvNeXt-Tiny model through a tailored preprocessing pipeline.
• To compare the proposed model against custom and state-of-the-art baselines.

Main Methods:

• An enhanced ConvNeXt-Tiny framework was developed, integrating a specialized preprocessing pipeline.
• The model was trained and evaluated on an RGB image dataset of 3,000 date palm leaf samples across four classes (Bug, Dubas, Healthy, Honey).
• Performance was benchmarked against CNN-Attention, ResNet13-Attention, ViT, ECA-Net, and standard ConvNeXt-Tiny.

Main Results:

• The preprocessing-augmented ConvNeXt-Tiny model achieved superior performance across all evaluated metrics: accuracy, precision, recall, and F1-score.
• The proposed framework significantly outperformed both custom-built and existing state-of-the-art models.
• The lightweight nature of the solution proved effective for scalable pest detection.

Conclusions:

• The enhanced ConvNeXt-Tiny framework offers a scalable and accurate solution for automated pest detection in precision agriculture.
• Tailored preprocessing is a key factor in improving feature quality and model performance.
• This approach addresses limitations of existing methods, enabling more efficient crop monitoring.