<p>Potato leaf diseases such as early blight and late blight pose significant threats to global crop yield, necessitating accurate and efficient detection methods. While deep convolutional neural networks (CNNs) have shown remarkable success in plant disease classification, their computational demands often hinder deployment on edge devices. To address this challenge, we propose LCE-Net (Lightweight CBAM-Enhanced EfficientNet), a novel architecture that integrates a lightweight attention module (Lite-CBAM) into the EfficientNet-B0 backbone. LCE-Net enhances feature extraction in spatial and channel dimensions with minimal computational overhead. We adopt a partial fine-tuning strategy and extensive data augmentation to improve generalization on limited datasets. Experimental results across two public potato leaf datasets demonstrate that LCE-Net achieves 100% classification accuracy, outperforming several models including Xception, InceptionV3, ResNet101, and MobileNetV2. Ablation studies confirm the critical role of attention mechanisms in boosting performance, while Grad-CAM visualizations verify model interpretability. These findings highlight LCE-Net as a highly accurate and lightweight solution suitable for real-time plant disease detection in smart agriculture.</p>