Scans

We first use the following bash code to aggregate images and convert pdf to png.

cd data/derived/
mkdir pdfs
ls ../raw/scans/*/*
cp ../raw/scans/*/* pdfs
mkdirs pngs
for file in $(ls pdfs/)
do
  pdftoppm pdfs/$file pngs/$file -png
done

We then use ImageJ (https://imagej.net/ij/download.html) with the following steps:

• Open virtual stack
• Type > 8-Bit
• Adjust > Threshold > Automatic
• Analyze Particles
  • Size: 1000-Infinity
  • Circularity: 0-1
  • Display results, clear results, exclude on edges, summarize, composite ROIs
  • with or without include holes depending on the question

The intermediate threshold images with detected are saved for manual check.

Code

la_ex <- read_csv("data/derived/area_excluded_summary.csv") %>% 
  select(Slice, `Total Area`) %>% 
  rename(slice = Slice, area_excluded = `Total Area`) %>% 
  separate(slice, c("vernacular", "tree", "leaf")) %>% 
  mutate(area_excluded = area_excluded / (1275*1755) * (21*29.7))
la_in <- read_csv("data/derived/area_included_summary.csv") %>% 
  select(Slice, `Total Area`) %>% 
  rename(slice = Slice, area_included = `Total Area`) %>% 
  separate(slice, c("vernacular", "tree", "leaf")) %>% 
  mutate(area_included = area_included / (1275*1755) * (21*29.7))
la_ex %>% 
  left_join(la_in) %>% 
  write_tsv("data/derived/la.tsv")

Code

read_tsv("data/derived/la.tsv") %>% 
  group_by(vernacular) %>% 
  mutate(la_s = mean(area_included)) %>% 
  ungroup() %>% 
  mutate(vernacular = fct_reorder(vernacular, la_s)) %>% 
  ggplot(aes(vernacular,
             area_included)) +
  geom_boxplot() +
  geom_jitter(aes(col = as.factor(tree)),
              width = 0.2, size = 3) + 
  theme_bw() +
  labs(x = "Vernacular name", y = "LA [ cm2 ]") +
  theme(axis.text.x = element_text(angle = 45, hjust = 1),
        legend.position = "none") +
  scale_y_log10()