In-Field Early Disease Recognition of Potato Late Blight Based on Deep Learning and Proximal Hyperspectral Imaging

Effective early detection of potato late blight (PLB) is an essential aspect cultivation. However, it a challenge to detect at stage in fields with conventional imaging approaches because the lack visual cues displayed canopy level. Hyperspectral can, capture spectral signals from wide range wavelengths also outside wavelengths. In this context, we propose deep learning classification architecture for hyperspectral images by combining 2D convolutional neural network (2D-CNN) and 3D-CNN cooperative attention networks (PLB-2D-3D-A). First, 2D-CNN are used extract rich space features, then mechanism AttentionBlock SE-ResNet emphasize salient features feature maps increase generalization ability model. The dataset built 15,360 (64x64x204), cropped 240 raw captured experimental field over 20 genotypes. accuracy test 2000 reached 0.739 full band 0.790 specific bands (492nm, 519nm, 560nm, 592nm, 717nm 765nm). This study shows encouraging result PLB proximal imaging.