TY  - JOUR
T1  - A Study on Meta-Learning Based Neural Architecture Search for Crop Disease Diagnosis in IoT Environments
AU  - Kim, Ji-Uk 
AU  - Lee, Hyun-Suk 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2025
DA  - 2025/1/1
DO  - 10.7840/kics.2025.50.9.1353
KW  - Meta-Learning
KW  - Neural Architecture Search
KW  - Crop Disease Diagnosis
KW  - IoT
AB  - In agriculture, accurate and efficient diagnosis of crop diseases is essential to improve agricultural productivity and quality. However, in real-world agricultural environments, there are challenges such as the lack of high-quality data and the limited memory and computational capacity of IoT devices. Therefore, in order to address crop disease diagnosis with deep learning, a lightweight model with a small size should be able to effectively examine crop disease with a small amount of training data. In this paper, we propose a crop disease diagnosis method that applies a combination of meta learning and neural architecture search to address these issues. The proposed method searches for a meta-model that can be generalized for various crop disease diagnosis tasks with only a small amount of data, in a search space consisting of lightweight models. Through experiments on real-world crop disease dataset, we demonstrate that the model trained with the proposed method achieves an accuracy improvement of more than 15.5% with 98.7% fewer parameters compared to the model in the related work. These results show that the proposed method is feasible for crop disease diagnosis under the limited conditions of real-world agricultural environments.