Ensemble Deep Learning Study for Pest Detection in Strawberry Plants

Abstract

This deep learning-based classification model was developed to recognize different types of pest infections in strawberry plants. The model aims to quickly identify pest symptoms, thus enabling efficient pest management in smart farming. This research uses an actual dataset containing images of strawberry leaves collected from smart farm trials. To expand the dataset, open data from platforms such as Kaggle were used, while images of infected leaves were obtained through web crawling with the help of Python libraries. The added data were converted to a uniform size, and PseudoLabeling was used to ensure stable learning on both training and testing datasets. The RegNet and EfficientNet models are selected as the main CNN-based models for iterative learning, with ensemble learning techniques to improve prediction accuracy. The proposed model aims to assist the early identification and treatment of pests on strawberry leaves during the early planting period, a crucial phase in the development of smart agriculture. It is hoped that this model can increase production in the agricultural industry and strengthen its competitiveness. Detecting early symptoms of plant diseases and pests is essential to prevent their development and minimize the damage caused. Although many methods have been developed using deep learning techniques, detecting early symptoms is still challenging due to the lack of datasets capable of training models against subtle changes in plants. Therefore, researchers built an automated data collection system to gather a large dataset of plant images and train ensemble models to detect diseases and pests of the target plants.