Automatic plant identification using transfer learning

Abstract

Plant identification is a widely researched area in the field of computer vision. Many attempts have been made to automate the process of plant identification using an image of a part of plant including flower, leaf and bark. Leaf has proven be the most reliable source of information. After exhaustive experiments, we chose to apply transfer learning to compare the feature extraction capabilities of VGG-16, Xception, MobileNetV2 Convolutional Neural Network (CNN) and DenseNet121 architectures to freely available Swedish, Flavia and MalayaKew leaf image datasets. Random Forest is used as classifier to identify the species of given leaf. The evaluations and comparisons of the specified feature extractor models are provided. DenseNet121 achieved maximum accuracy of 100%, 99% and 92.4% respectively in the three datasets.