Abstract
Lack of annotated data for training of deep learning systems is a challenge for many visual recognition tasks. This is especially true for domain-specific applications, such as plant detection and recognition, where the annotation process can be both time-consuming and error-prone. Generative models can be used to alleviate this issue by producing artificial data that mimic properties of real data. This work presents a semi-supervised generative adversarial network (GAN) model to produce artificial samples of plant seedlings. By applying the semi-supervised approach, we are able to produce visually distinct samples for nine unique plant species using a single GAN model, while still maintaining a relatively high visual variance in the produced samples for each species. Additionally, we are able to control the appearance of the generated samples with respect to rotation and size through a set of latent variables, despite these not being annotated features in the training data. The generated samples resemble the intended species with an average recognition accuracy of ∼64.3%, evaluated using an external state-of-the-art plant seedling classification model. Additionally, we explore the potential of using the GAN model’s discriminator as a quality assessment tool to remove poor representations of plant seedlings from the artificial samples.
Highlights
Machine learning, and in particular deep learning, have become increasingly popular in recent years
The WacGAN-info model could be used to generate an additional test-set of artificial samples for the class discriminability test on the ResNet-101 classifier. Both the WacGAN-info model and the external ResNet-101 classifier were trained on the segmented plant seedling dataset by Giselsson et al [27]
The results show that the WacGAN-info model is capable of producing artificial samples that resemble real plant seedlings
Summary
In particular deep learning, have become increasingly popular in recent years. Deep learning algorithms are still primarily trained using supervised learning, which requires a substantial amount of annotated training data. There exist many publicly available datasets for training deep learning systems [1]; when it comes to domain-specific applications such as plant detection and recognition, the availability of data is significantly reduced [2]. While GAN models can be used in many different generative applications, they are most often used for generating artificial images. A strength of the GAN models is that they can be trained using an unsupervised and/or supervised end-to-end learning approach [4–6]. Previous research has shown that artificial GAN samples can be used as data augmentation to increase the performance of several visual recognition tasks [7–10]
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have