PlantDefender: a unified deep learning approach for maize disease and pest detection

Abstract

Maize crops face severe threats from various diseases and pests, significantly affecting crop quality and yield. Prompt and accurate diagnosis is essential to minimize agricultural losses. Existing models have demonstrated limitations in handling complex backgrounds, inter- and intra-class variations, and inefficient feature extraction, which complicate the recognition of maize diseases and pests. This study proposes a unified deep learning model, termed PlantDefender, for the simultaneous recognition of maize diseases and pests. PlantDefender employs an improved Faster R-CNN architecture with a refined InceptionResNet-V2 as its backbone. The conventional InceptionResNet-V2 is modified by replacing the ReLU activation function with Swish to enhance feature extraction. The extracted features are subsequently processed through the Region Proposal Network (RPN), ROI Align, regression, and classification modules for accurate detection and recognition. The model is evaluated using the IP102 dataset for maize pest recognition and the CD&S dataset for maize leaf disease recognition. PlantDefender achieves an accuracy of 78.67% on the IP102 dataset and 98.44% on the CD&S dataset. The results demonstrate effective localization and classification performance across both maize pest and disease recognition tasks. The experimental results indicate that the proposed PlantDefender framework can effectively address the challenges associated with the recognition of maize diseases and pests. By integrating the refined InceptionResNet-V2 backbone with Faster R-CNN, the proposed unified model provides a promising approach for automated and accurate maize crop protection.