A novel plant type, leaf disease and severity identification framework using CNN and transformer with multi-label...

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Plant diseases pose a significant threat to agriculture, necessitating accurate and timely identification for effective management.
• Current deep learning methods for leaf disease identification often lead to complex models or numerous categories by treating plant type, disease, and severity separately or combined.
• A simplified and integrated approach is needed to improve the efficiency and accuracy of leaf disease identification systems.

Purpose of the Study:

• To propose a novel leaf disease identification network (LDI-NET) capable of simultaneously identifying plant type, leaf disease, and severity.
• To develop a single-branch deep learning model that avoids increasing category numbers or employing complex network structures.
• To enhance the accuracy and efficiency of plant disease detection through an integrated multi-label identification approach.

Main Methods:

• The LDI-NET employs a multi-label approach with three core modules: a feature tokenizer, a token encoder, and a multi-label decoder.
• The feature tokenizer module integrates convolutional neural networks and transformers to extract both local and global contextual features.
• The token encoder establishes relationships between plant type, disease, and severity, while the multi-label decoder fuses features for simultaneous identification.

Main Results:

• The proposed LDI-NET successfully identifies plant type, leaf disease, and severity concurrently within a single, straightforward network branch.
• Experimental results demonstrate that LDI-NET outperforms existing prevalent methods on the AI challenger 2018 dataset.
• The model effectively leverages shallow and deep features through its residual structure in the decoder module.

Conclusions:

• LDI-NET offers a novel and effective solution for simultaneous plant type, leaf disease, and severity identification.
• The multi-label approach integrated into a single network branch simplifies model architecture and improves identification performance.
• This research contributes to advancing automated plant disease diagnostics, crucial for agricultural sustainability and food security.