Plant-seedling classification using transfer learning-based deep convolutional neural networks

Abstract

The ever-growing population of this world needs more food production every year. The loss caused in crops due to weeds is a major issue for the upcoming years. This issue has attracted the attention of many researchers working in the field of agriculture. There have been many attempts to solve the problem by using image classification techniques. These techniques are attracting researchers because they can prevent the use of herbicides in the fields for controlling weed invasion, reducing the amount of time required for weed control methods. This article presents use of images and deep learning-based approach for classifying weeds and crops into their respective classes. In this paper, five pre-trained convolution neural networks (CNN), namely ResNet50, VGG16, VGG19, Xception, and MobileNetV2, have been used to classify weed and crop into their respective classes. The experiments have been done on V2 plant seedling classification dataset. Amongst these five models, ResNet50 gave the best results with 95.23% testing accuracy.