A Diseased Rice Plant Detection Method Based on Transfer Learning Technique

EasyChair Preprint 15410

Abstract

Early detection of diseases in rice plants is crucial for ensuring crop health and optimizing yield. Traditional methods of disease identification are time-consuming and require expert knowledge, which limits their practical application in large-scale agriculture. Using drones equipped with cameras and computer vision systems will be an effective solution. This paper presents a rice plant disease detection method using transfer learning techniques on images captured from the view of drones. We leverage EfficientNet B0 pre-trained to extract features from images of rice field regions and fine-tune these networks to distinguish diseased rice from disease-free rice. Our dataset comprises high-resolution images of healthy rice fields and regions affected by common diseases such as bacterial blight, brown spots, and leaf blasts. Experimental results demonstrate that the proposed method outperforms popular image classification methods based on CNN for Rice Plant Disease Detection in our dataset in terms of accuracy and inference time.

Keyphrases: Diseased Rice Plant Detection, Transfer Learning, deep learning

@booklet{EasyChair:15410,
  author    = {Doan-Tung Duong and My-Trinh Ha and Van-Toi Nguyen},
  title     = {A Diseased Rice Plant Detection Method Based on Transfer Learning Technique},
  howpublished = {EasyChair Preprint 15410},
  year      = {EasyChair, 2024}}