Paraleafnet: A Lightweight Parallel Cnn for Efficient Plant Disease Identification in Precision Agriculture

Abstract

Effective plant disease detection is vital for sustainable agriculture; however, the computational demands of many deep learning frameworks make them impractical for use in low-resource settings. This study proposes ParaLeafNet, a streamlined Parallel Convolutional Neural Network (CNN) that merges MobileNetV2 and MobileNetV3Small with a Squeeze-and-Excitation (SE) Attention mechanism to improve feature extraction. Tailored for edge applications, ParaLeafNet underwent optimization via TensorFlow Lite and was tested on the PlantVillage dataset, with ablation studies examining the roles of its parallel design and attention system. ParaLeafNet outperformed standard CNN models in plant disease classification, providing both precision and computational efficiency. Visualization techniques confirmed its ability to pinpoint critical disease markers, boosting its utility for real-world scenarios. ParaLeafNet delivers a powerful deep learning solution for real-time plant disease monitoring, fostering sustainable farming practices by enabling farmers to tackle challenges early, curb losses, and advance precision agriculture. Its lightweight architecture ensures compatibility with resource-constrained devices, supporting broader food security goals. Future work will prioritize diverse real-world datasets and enhancements for ultra-low-power systems

Code and data availability

The paper uses the public PlantVillage dataset, but that is cited prior work (ref [31]), not a paper-specific asset. No author code, trained model checkpoints, or data availability statement with a public URL appears in the supplied blocks.

No evidence-backed public reproduction asset is currently recorded.