Updated: Nov 1, 2021
This summer, our R&D teams have focused on getting Verifli ready to handle a major ramp up scale in 2022. The software team has prioritized improving the Verifli dashboard user experience, debugging the mobile app and designing a program to speed up our image quality control process. Meanwhile, the data science team worked on streamlining the image analysis pipeline, adapting the Verifli predictive model to warm weather and processing data for pilot tests.
In preparation to deploy Verifli to new crop pollination events that take place in the summer months, we needed to adjust our approach to data collection. In past years, we’ve strictly sought to collect data in conditions that mirror typical February weather in California’s Central Valley. Now that the Verifli model has exceeded our accuracy target in temperatures around 45-70 degrees, the data collection team has been tasked with gathering data in warmer climates. We’ve completed 4 data collection trips so far, visiting some of the country’s most brutally hot areas, where shady spots to rest come at a premium.
The data science team has juggled loads of new work over the summer to gear up for expansion in 2022. As we start to grade hives for other pollination events occurring in the spring and summer, increasing efficiency and capacity of the image analysis pipeline is key to successful growth.
After identifying major bottlenecks, we automated the slowest and most manual-intensive processes in order to reduce the personnel hours needed to operate the pipeline. Not only will this benefit our customers by generating faster hive strength results, it’ll free up time in 2022 for the data science team to spend on building new things, rather than spending hours each day monitoring the pipeline.
Meanwhile, we’ve been tuning the hive strength model to deliver accurate results in warm weather. In weather below about 60 degrees, bees cluster together tightly to insulate the brood and regulate colony temperature. When the weather gets warmer, bees begin to spread out more within the hive to prevent overheating. Thanks to the efforts of the data collection team, we’ve trained the hive strength model on warm weather bee behavior, resulting in accuracy that matches our original model used in almond pollination.
As for software development, the theme is more of the same: efficiency and scale. The software and data science teams have collaborated to update the back-end software architecture to support the new automations in the image analysis pipeline. They also joined forces to build a tool designed to speed up image segmentation quality control—a step in the pipeline that must be performed manually.
A new graph on the Verifli dashboard Reporting page, showing daily weather forecast and projected bee flight hours. Check it out for yourself on our demo dashboard.
With pilots underway around the globe, the software team ran tests to ensure that the mobile app and web dashboard function properly in new countries. We also added new features to the Verifli dashboard to give users additional information to evaluate pollination. Among other new data on the dashboard, we designed a custom formula that pairs the upcoming weather forecast with your hive strength results to estimate the quality of pollination our customers can expect to receive.
We’ve made impressive progress in the R&D department so far this year in spite of the increased workload resulting from nearly a dozen pilot tests. Looking ahead to the final months of 2021, our R&D team will remain focused on improving efficiency and preparing the pipeline to handle significantly more images. With a major leap in scale on the horizon, we’re working hard to prepare our technology to handle the issues we anticipate to encounter.