Edge-Cloud Remote Sensing Plant Disease Detection via Multilayer Hyperbolic Lyrebird Convolutional Attention Networks

Plant disease detection (PDD) is essential for improving crop productivity, ensuring food security, and supporting sustainable agriculture. However, real-time and accurate detection is challenging due to complex lesion patterns, diverse crop environments, and large-scale data. To address these issues, this paper presents a novel edge–cloud framework called Fully Multilayer Hyperbolic Edge Lyrebird Convolutional Attention Network (Fully-MHE-LBCAN) for intelligent plant disease detection. The framework integrates Shape-Aware Mesh Normal Filtering (SAMNF) for effective noise removal and lesion boundary preservation, followed by the Scale-Aware Modulation Meet Transformer (SAMMT) to extract multi-scale spatial, structural, and contextual features from images, videos, and numerical crop data. A Fully Hyperbolic Convolutional Neural Network (FHCNN) models hierarchical disease representations in hyperbolic space, while a Multilayer Edge Attention Network (MEAN) enhances lesion-focused learning and supports efficient edge–cloud collaboration. Additionally, the Lyrebird Optimization Algorithm (LOA) optimizes network parameters, enabling improved stability and reduced computational complexity. Extensive experiments on multi-modal remote-sensing datasets involving sugarcane, blueberry, cotton, and cherry crops show that Fully-MHE-LBCAN achieves 99.9% accuracy with a 0.1% error rate, outperforming existing methods in accuracy. The results confirm that the proposed framework is scalable, reliable, and well-suited for real-time large-scale plant disease detection in precision agriculture.