Early Crop Disease Identification Using Multi-fork Tree Networks and Microbial Data Intelligence

In the early stages of crop disease, timely acquisition of information about crop diseases, determination of the causes and severity of infection, and targeted treatment are essential for preventing a decline in crop yield caused by disease spread. To address the issue of low accuracy in traditional deep learning networks for early crop disease identification, we propose an improved attention mechanism-based multi-fork tree network method. This method combines the attention mechanism with a residual network to recalibrate disease feature maps, resulting in SMLP_Res (Squeeze-Multi-layer Perceptron ResNet). Additionally, we extend the high-feature extraction-capable SMLP_ResNet (Squeeze-Multi-Layer Perceptron ResNet) network with a multi-fork tree structure, simplifying the task of early crop disease identification and effectively extracting early disease features. In our experiments, we use two datasets, Plant Village and AI Challenger 2018, to train and validate three network models: 18-layer ResNet, SE_ResNet, and SMLP_ResNet, as well as their equivalent multi-fork tree structure models, to assess the impact of SMLP_Res and the multi-fork tree structure on crop disease identification models. The experimental analysis shows that the three network models, 18-layer ResNet, SE_ResNet, and SMLP_ResNet, all achieve an accuracy rate of over 99% in disease identification on the Plant Village dataset, where disease features are more pronounced. However, their accuracy rates on the early disease dataset AI Challenger 2018 do not exceed 87%. SMLP_ResNet, due to the inclusion of the SMLP_Res module, provides more comprehensive feature extraction for crop disease data, resulting in better detection performance. Among the three early disease identification models with multi-fork tree structures, all models show significant improvements in accuracy on the AI Challenger 2018 dataset. The multi-fork tree SMLP_ResNet outperforms the other two models, achieving the best performance with a cherry early disease identification accuracy rate of 99.13%. The proposed multi-fork tree SMLP_ResNet crop early disease identification model simplifies the recognition task, suppresses noise transmission, and achieves high accuracy.