We took pictures of 10 types of vegetables from 4 stores over the course of 5 days, for a total of 200 pictures (you can download the full data here). Out of these pictures, we selected a few sets for more detailed analysis.
Day 1 vs Day 5 Analysis
To compare the effectiveness of each post-processing method (Infrablue, Exposed-Film, and No IR-Blocker), we chose to look at the results of the NDVI images of broccoli from the first and last day of shooting.
As a comparison of the clarity of the post-processing methods, this analysis showed the Infrablue pictures as significantly lower quality. It is unclear whether this is due to the quality of the images themselves (blurriness, lighting) or due to the post-processing method. The two multi-image post-processing methods both produced very clear images, but the ranges represented in the Exposed Film images seem more realistic and the images are easier to generate (no color balancing needed).
Unfortunately, the data itself proved inconclusive. There were no significant changes in photosynthesis over the 5-day period, nor were there obvious differences in photosynthetic strength between the produce from different locations.
Once we determined the clearest post-processing method, we applied it to other vegetables and mapped their progression over the course of the 5 days into time-lapses.
The kale very noticeably wilted after the 5 day period, yet in the NDVI spectrum there were either small or inconsistent differences.
The cilantro follows a similar trajectory, with strong visual-spectrum changes coupled with little to no NDVI changes.
We were unable to find clear differentiation on any of the axes we tested (produce location, vegetable type, decay over time); it appears that using near-infrared photography to glean meaningful data about local produce is not feasible. We don’t know the deeper connections between photosynthetic activity and produce health, so attempting to draw other conclusions based off the correlations (or lack thereof) we see in our data would not be useful.
We compiled select grids of the pictures we would take on a daily basis to show the unaltered results of our cameras. Although we derived the NDVI images from these pictures, they underwent significant transformations before they could be used (color correction, perspective transforms).
Continue on to our reflections.