A Lightweight Maize Pest and Disease Recognition Method Based on an Improved ShuffleNetV2 in Field Environments

Abstract

Abstract Agricultural pest and disease monitoring plays a vital role in ensuring crop productivity, reducing pesticide consumption, and promoting sustainable agricultural development. Although deep learning techniques have achieved remarkable success in plant health diagnosis, many existing models remain computationally intensive and are difficult to deploy on resource-constrained edge devices used in practical agricultural environments. To address these challenges, this study proposes a practical lightweight deep learning framework based on an improved ShuffleNetV2 architecture for real-time maize pest and disease recognition under complex field conditions.The proposed model incorporates the Ghost module to reduce redundant feature generation, the Efficient Channel Attention (ECA) mechanism to enhance feature representation, and the HardSwish activation function to improve nonlinear learning capability while maintaining computational efficiency. Extensive experiments were conducted on a maize pest and disease dataset containing multiple disease and pest categories collected under natural field conditions. Experimental results demonstrate that the proposed model achieves superior recognition accuracy while significantly reducing model parameters and computational complexity compared with several mainstream lightweight convolutional neural networks.The results show that the proposed method achieves an accuracy of 93.00%, a recall of 92.76%, and an F1-score of 92.42%, while maintaining extremely low computational cost (0.03 GFLOPs) and model size (1.16 MB). Furthermore, the proposed model was successfully deployed on a Raspberry Pi platform, demonstrating excellent real-time inference capability and low computational resource consumption. The framework is suitable for practical agricultural applications, including intelligent crop monitoring, UAV-assisted field inspection, and mobile diagnostic systems. By enabling rapid and accurate in-field identification of maize pests and diseases, the proposed approach supports timely crop protection decisions, reduces unnecessary pesticide application, and contributes to sustainable agriculture through practical edge-AI deployment.

Plant phenotyping relevance

トウモロコシの病害状態を画像から認識する軽量深層学習法の開発・評価が中心であり、植物病害フェノタイプの取得手法に該当する。

Code and data availability

The paper's maize pest/disease image dataset (19,451 images compiled from PlantVillage plus web-crawled images) is not publicly deposited; the Data Availability Statement says raw data are available only on request. No author code, models, or public repository URLs are provided; PlantVillage itself is a cited prior公共资源

No evidence-backed public reproduction asset is currently recorded.