DExNet: Combining Observations of Domain Adapted Critics for Leaf Disease Classification with Limited Data

2025-06-22

AI Summary

To address the challenge of few-shot plant leaf disease classification, this paper proposes DExNet: a framework that integrates nine domain-adaptive CNN expert models to extract cross-domain robust features, and innovatively introduces a Bi-LSTM-driven dynamic feature fusion mechanism for temporal modeling and discriminative aggregation of multi-source observations. The method significantly reduces reliance on labeled data—achieving 89.06%–94.07% accuracy on the PlantVillage tomato subset under 5- to 15-shot settings, and 98.09% under 80-shot—reaching near state-of-the-art performance using only 6% of the training data. Its core contribution lies in the first integration of domain adaptation with a multi-expert Bi-LSTM architecture for few-shot plant disease recognition, demonstrating strong generalization and stability across both cross-domain and mixed-scenario benchmarks.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

While deep learning-based architectures have been widely used for correctly detecting and classifying plant diseases, they require large-scale datasets to learn generalized features and achieve state-of-the-art performance. This poses a challenge for such models to obtain satisfactory performance in classifying leaf diseases with limited samples. This work proposes a few-shot learning framework, Domain-adapted Expert Network (DExNet), for plant disease classification that compensates for the lack of sufficient training data by combining observations of a number of expert critics. It starts with extracting the feature embeddings as 'observations' from nine 'critics' that are state-of-the-art pre-trained CNN-based architectures. These critics are 'domain adapted' using a publicly available leaf disease dataset having no overlapping classes with the specific downstream task of interest. The observations are then passed to the 'Feature Fusion Block' and finally to a classifier network consisting of Bi-LSTM layers. The proposed pipeline is evaluated on the 10 classes of tomato leaf images from the PlantVillage dataset, achieving promising accuracies of 89.06%, 92.46%, and 94.07%, respectively, for 5-shot, 10-shot, and 15-shot classification. Furthermore, an accuracy of 98.09+-0.7% has been achieved in 80-shot classification, which is only 1.2% less than state-of-the-art, allowing a 94.5% reduction in the training data requirement. The proposed pipeline also outperforms existing works on leaf disease classification with limited data in both laboratory and real-life conditions in single-domain, mixed-domain, and cross-domain scenarios.

Problem

• Classifying leaf diseases with limited training data
• Combining domain-adapted critics for few-shot learning
• Reducing data requirements while maintaining high accuracy

Innovation

• Few-shot learning with domain-adapted expert critics
• Feature fusion from multiple pre-trained CNNs
• Bi-LSTM classifier for enhanced accuracy