Authors: Nathan Jones, Carter Noh, Douglas Cook
Mentors: Douglas Cook
Insitution: Brigham Young University
Neural networks are used to identify specific objects in a picture and are often used in robotics to allow robots to identify objects through a camera. They can be used in agriculture to allow machines to identify plants to harvest and cultivate. The preparation of the neural network model involves taking thousands of pictures in a variety of situations. Networks with a large quantity of pictures in a large variety of angles, lighting, environments, etc. have a better chance of identifying plants in any situation. Our lab needed a device for rapid data collection that could be placed in a field to automate the process of taking pictures of crops.
We used a gantry system that was controlled through three stepper motors, one for the x-axis and the other two for the y-axis. Two cameras were placed on the head of the gantry system and were driven through a 40” x 40” area. Each camera was connected by a double ball head arm which can point a camera in almost any direction. The camera arms were screwed into a plate with a 5 x 5 grid of bolt holes; these holes and the double ball head arms gave us control over the angle and distance of each photo to increase the variety of our data set.
In October 2023, our rapid data collection device was tested in a field and was able to capture over 50,000 photographs of saffron flowers at a variety of angles, lighting, and distances. The results of our device were promising and we have some improvements that we plan on making. We anticipate that with these improvements, next saffron harvest we will be able to increase the number and variety of pictures to improve our dataset.