Tomato leaf disease recognition based on multi-task distillation learning

Abstract

Introduction: Tomato leaf diseases can cause major yield and quality losses. Computer vision techniques for automated disease recognition show promise but face challenges like symptom variations, limited labeled data, and model complexity.

Methods: Prior works explored hand-crafted and deep learning features for tomato disease classification and multi-task severity prediction, but did not sufficiently exploit the shared and unique knowledge between these tasks. We present a novel multi-task distillation learning (MTDL) framework for comprehensive diagnosis of tomato leaf diseases. It employs knowledge disentanglement, mutual learning, and knowledge integration through a multi-stage strategy to leverage the complementary nature of classification and severity prediction.

Results: Experiments show our framework improves performance while reducing model complexity. The MTDL-optimized EfficientNet outperforms single-task ResNet101 in classification accuracy by 0.68% and severity estimation by 1.52%, using only 9.46% of its parameters.

Discussion: The findings demonstrate the practical potential of our framework for intelligent agriculture applications.

Keywords: disease classification; knowledge distillation; multi-task learning; severity prediction; tomato leaf diseases.