Fig 1.
Moths were filmed using a Vision Research Phantom camera and lit from behind using near-IR LED lamps and an optical diffuser to maximize contrast. This example image has been labeled manually. Moths fed either while flying or while standing on a perch (not shown).
Fig 2.
Test performance of object-locating neural networks.
A: Meniscus network performance. B: Proboscis network performance. Histogram of intersection-over-union measurements for meniscus-tracking and proboscis-tracking neural networks on a test set randomly selected from the frames in the data set.
Fig 3.
Measured nectar volume intake measurements approximate daily ingested nectar mass.
Scatter plot of daily intake volume measured from neural network analysis of video data against daily mass change. Each point shows the total daily nectar volume and mass change for an individual moth. Mass intake and volume intake follow a significant (R = 0.7, p = 0.0002) linear relationship, with a slope less than 1. This is expected, because the mass measurement encompasses the entirety of a feeding bout whereas the video measurement captures an unknown portion of it. Regression line is in solid blue with a shaded 95% confidence interval, a 1-to-1 line is plotted in black dashes.
Fig 4.
Effect of proboscis extension on feeding rate.
A violin plot of drinking rate against feeding behavior, with each datapoint being one recording. Recordings with a curled proboscis tend to have significantly (p = 0.0007) lower drinking rates than flying feeding bouts with an extended proboscis by a Mann-Whitney U-rank test. When the moth is flying and feeding, only extended proboscis postures were observed.
Table 1.
Recording and behavior differences.
Fig 5.
Nectar ingestion rate is independent of proboscis submergence.
Scatterplot with superimposed linear model of per-recording median nectar ingestion rate against per-recording median proboscis submergence, with shaded 95% confidence interval. Each point shows the median submergence depth and median drinking rate for one recording. Points are colored by the moth’s ID and styled by whether the moth is perching (circle) or flying (x) in that recording.