English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In this paper we report on improved part segmentation performance usingconvolutional neural networks to reduce the dependency on the large amount ofmanually annotated empirical images. This was achieved by optimising the visualrealism of synthetic agricultural images.In Part I, a cycle consistentgenerative adversarial network was applied to synthetic and empirical imageswith the objective to generate more realistic synthetic images by translatingthem to the empirical domain. We first hypothesise and confirm that plant partimage features such as color and texture become more similar to the empiricaldomain after translation of the synthetic images.Results confirm this with animproved mean color distribution correlation with the empirical data prior of0.62 and post translation of 0.90. Furthermore, the mean image features ofcontrast, homogeneity, energy and entropy moved closer to the empirical mean,post translation. In Part II, 7 experiments were performed using convolutionalneural networks with different combinations of synthetic, synthetic translatedto empirical and empirical images. We hypothesised that the translated imagescan be used for (i) improved learning of empirical images, and (ii) thatlearning without any fine-tuning with empirical images is improved bybootstrapping with translated images over bootstrapping with synthetic images.Results confirm our second and third hypotheses. First a maximumintersection-over-union performance was achieved of 0.52 when bootstrappingwith translated images and fine-tuning with empirical images; an 8% increasecompared to only using synthetic images. Second, training without any empiricalfine-tuning resulted in an average IOU of 0.31; a 55% performance increase overprevious methods that only used synthetic images.

Improved Part Segmentation Performance by Optimising Realism of Synthetic Images using Cycle Generative Adversarial Networks