Yesterday, Fly Bye attended the Final Design Review in the Reitz Grand Ballroom. It was an awesome experience being able to show off all the hard work our team accomplished this semester. We were able to showcase our poster, our video, and our presentation to other teams and visitors. Our Liaison attended FDR and provided a large bug net to see our project in action. It was nice having our parents and friends attend the showcase.

It is a bittersweet moment for the team, as this is the last time we will be presenting a project together and wear our shirts. Four out of the five members will graduate this year, so this is one of the last events most of Fly Bye will attend as students.

We are excited to see what will happen with the project in the future!

As a side note, the food at FDR was delicious! It was a nice treat for the team after the Panelist of Self-Regulated Engineers.

Fly Bye Video

This week, Fly Bye presented a preliminary FDR presentation to students and faculty coaches in the PR-FDR event. The image below is a plot that we presented during the PR-FDR event. We received detailed feedback including recommendations on how to improve aspects of the FDR presentation for the actual event on Tuesday April 25th. We also completed the poster and video required for the FDR event next week. Fly Bye is looking forward to delivering our prototype to the USDA-ARS during the FDR event and showing the public the progress that we have made on the image-based prototype throughout the two-semester IPPD program. As the IPPD program comes to a close and the semester ends, we are reflecting on the work that we have accomplished and are working diligently to document the project so further groups can pick up on this work.

This week, we fine-tuned our machine-learning model and our housing design. Our model achieved an average precision (AP) score of around 50%, which is on the order of the AP scores of other top-quality models! We are still fine-tuning the parameters to see if a slightly better AP score can be achieved. However, as is, the model is very capable of detecting invasive insects with decent accuracy. We also wrote a program to concatenate images with detected insects with the original image. The image below is an example of this feature: the left one shows a successfully detected Caribbean Fruit Fly and the right one shows the original image.

The end is approaching! With FDR only 20 days away team Fly Bye is making the final push to polish the image based prototype. The PCB for the prototype is being soldered together and the Machine Learning model is continuing to be trained. The team has also split up responsibilities for Final Design Report and is excited to had over the final product to the Liaison at the USDA. Along with the report, the team is working on a project poster to present at the Final Design Showcase. Alyssa is also working on editing a video that will explain the purpose of the project.

This week, team Fly Bye has been hard at work to finalize the project deliverables in order to have a robust prototype for Prototype Inspection Day. In this endeavor, they were able to establish a housing scheme that will account for the flashing of the camera, establish a pipeline for the machine learning model to run on and detect invasive insects, establish a PCB, and continue working toward improving the model accuracy. Although the finish line is near, it is not here yet, and team Fly Bye is aiming to finish the race strong in the next few weeks as they finalize their product and accompanying documentation.

This week Fly Bye is back from spring break! The team has been working this week to prepare for Prototype Inspection Day, which is next Tuesday. Over the next few weeks, Fly Bye is beginning to wrap up the project and fix up any loose ends.

This past week Fly Bye designed and ordered a printed circuit board that will be used in the final design of the image-based trap. This will make the system look more professional and help the electronics fit more securely in the trap.

Fly Bye was able to make progress in the image processing. A Google Apps script will process any images uploaded to Google Drive and immediately send them to the machine learning model, which will identify any fly images.

Prior to spring break, Fly Bye was able to collect 600 images of flies. This week, Fly Bye wrote a script to augment the data by flipping all the existing images. This means now the model will have 1200 images. At this point, almost all the images have been annotated.

The next few images are of the data collection process and the live flies the USDA sent the team.

Next week, Fly Bye plans to present updates at the Prototype Inspection Day, continue to collect data, and solder the components onto the printed circuit board.

This final week before spring break, Fly Bye worked diligently to finalize a prototype which will be tested over break in Miami. Existing images taken for data collection were labeled and new images were collected. Specifically, 150 raw images were collected containing Mediterranean fruit flies, and 70 images were gathered including Oriental and Caribbean fruit flies. Fly Bye now has now gathered a total of 600 raw images for the training dataset. After augmenting this data using varying methods, the dataset size will be significantly increased and the model accuracy will ideally increase proportionally.

Additionally, Fly Bye worked this week to iterate on components of the prototype including housing design and color, circuitry components, and the program that transfers the image from the device to the cloud. Specifically, a new housing will be printed in a white color, with the goal of this being more compatible with the fruit flies of interest. Circuitry components are being integrated into a printed circuit board which will be placed in the small electronics housing above the sticky trap platform. Overall, Fly Bye is excited to deliver a preliminary prototype for testing this spring break.

This week, Fly Bye reached several major milestones. First, we finally fixed the data transfer issue with our image-based prototype! After deciding to process data remotely, we returned to the issue of how to transfer more than 64 KB to Google Drive. On Wednesday, we fixed the issue and can now send images of several hundred KB. In addition, we received PCBs for the infrared prototype and began assembling them. We also took 300 pictures this week, reaching a total of 370 total images to train our convolutional neural network for the image-based prototype. An example of a high-quality image automatically sent to Google Drive is attached.

This week with Team Fly Bye, QRB 2 is rapidly approaching and the team is making great progress. Team work meetings have become more frequent as data collection and prototyping are at the top of our to do list. After a conversation with our Liaison the team has decided to focus its efforts primarily on the camera prototype to ensure the best results of it. While some work will still be done on the Infrared prototype when the PCBs are delivered most of the energy will be placed on making the best camera prototype possible. This week we ran into Team Monster Gator twice in the IPPD lab and have become fast friends! Alyssa also made some team shirts that you can see griffin holding below!