Classification of Vegetable Plant Pests using Deep Transfer Learning

Abstract

Sri Lankan farmers face several issues and among them crop loss due to insect pest infestation is the major hurdle. Several approaches have been proposed to detect the pest class by using computer vision and machine learning techniques. These approaches gain the classification knowledge by using a large number of pest image samples, despite it being time consuming. In this paper, we propose an approach to detect the vegetable plant pests in Sri Lanka. To our best knowledge, no previous studies were conducted to detect pest classes in Sri Lankan vegetables. We have used the deep transfer learning technique to train the classification model with fewer number of samples as they are able to transfer the learned knowledge from one domain to another. Raw images of ten vegetable pest classes were collected and then a database was constructed. The VGG16 and InceptionV3 Convolutional Neural Network (CNN) pre-trained models were used to transfer the classification knowledge and they showed the accuracy of 97.5% and 99.8 %, respectively.